See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/348430431

Multiple comparison test by Tukey's honestly significant difference (HSD): Do

the confident level control type I error

Article in International Journal of Applied Mathematics and Statistics · January 2021

DOI: 10.22271/maths.2021.v6.i1a.636

CITATIONS READS

41 9,189

4 authors, including:

Abikesh Prasad Kumar Mahapatra Bibhuti Bhusan Mohapatra

Sun Pharmaceutical Industries DPMishra Institute for Advance Studies in Special Education &Technology
18 PUBLICATIONS 121 CITATIONS 3 PUBLICATIONS 48 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Abikesh Prasad Kumar Mahapatra on 13 January 2021.

The user has requested enhancement of the downloaded file.

 International Journal of Statistics and Applied Mathematics 2021; 6(1): 59-65

ISSN: 2456-1452
Maths 2021; 6(1): 59-65
© 2021 Stats & Maths Multiple comparison test by Tukey’s honestly
www.mathsjournal.com
Received: 17-11-2020 significant difference (HSD): Do the confident level
Accepted: 29-12-2020
control type I error
Anita Nanda
DPMIASSE&T, Special BEd.
College, Kodola, Ganjam, Anita Nanda, Dr. Bibhuti Bhusan Mohapatra, Abikesh Prasada Kumar
Odisha, India
Mahapatra, Abiresh Prasad Kumar Mahapatra and Abinash Prasad
Dr. Bibhuti Bhusan Mohapatra Kumar Mahapatra
DPMIASSE&T, Special BEd.
College, Kodola, Ganjam, DOI: https://doi.org/10.22271/maths.2021.v6.i1a.636
Odisha, India

Abikesh Prasada Kumar Abstract

Mahapatra Examining a huge amount of data is a typical issue in any research process. However, different statistical
OPJS University, School of processes and techniques play essential role to derive a meaningful conclusion from the presented
Pharmacy, Churu, Rajasthan, enormous data. Control of type I error is highly essential for a researcher or statistician while dealing
India with comparisons test with more than two variables. Multiple testing statistical tests provides a structural
system and minimizes the error rate by helping to derive meaningful accurate conclusions. Among the
Abiresh Prasad Kumar different multiple test procedures Tukey's honestly significant difference test (Tukey's HSD) is most
Mahapatra common and popular techniques. The main objective of this study was to explore how significantly
BALVIKAS, B.A.M.M Institute selection of confidence level or error rate can affect the rate of committing type I error while drawing
for mental retarded, Kodola, conclusion. The effect of committing type I error with selection of confidence level or error rate was
Ganjam, Odisha, India explored with citing suitable case study in a special education setting. The case study focuses cognitive
effect of music on growth and enhancement of the motor behavior (running, jumping and sliding) of
Abinash Prasad Kumar children with mild intellectual disability enrolled in the special school setting. ANOVA test was
Mahapatra performed and the significance of selection of individual confidence level and simultaneous confidence
North Orissa University, level) in Tukey’s HSD test was described.
Department of computer science,
Baripada, Odisha, India Keywords: Experimental design, P value, significance level, Tukey’s HSD, statistical analysis, type I
error, ANOVA, special education

1. Introduction
A common problem inside the sciences and research is to examine numerous treatments to
figure out which, if any, produce a superior result [1]. A crucial and important tool which helps
the researchers to carry out studies on more than one experimental groups and control groups
is the Analysis of Variance (ANOVA).The analysis of variance (ANOVA) gives a significant
result, which identified the difference between the groups more probably it indicates that at
least one group differs from the other groups. However, the test don’t inform on the pattern of
differences between the means [2].
The major setback associated with ANOVA is that it only compares the means between groups
and the results concluded if all group means under consideration are equal or not, but does not
categorize the exact information such as which particular pairs of means are significant. In
order to get the accurate and specific pattern of difference between means, the ANOVA is
often equipped with specific test specific compares the two means between the pairs or groups
(the called pair wise comparisons) [3, 4].
Multiple comparisons of means allow to examine which means are different and to estimate by
how much they are different. The statistical difference between means can be assessed by
using statistical tools operates on fundamental concept of statistics using set of confidence
Corresponding Author: intervals or hypothesis tests or both [5, 6]. These types of pair wise comparison tests are called
Anita Nanda multiple comparison techniques and the most frequently technique was developed by Tukey
DPMIASSE&T, Special BEd. and named as the honestly significant difference (HSD) test. The main idea of the HSD is
College, Kodola, Ganjam,
works on the principle of computation of the honestly significant difference between two
Odisha, India
~59~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

means using a statistical process. The process gives the exact 2. The Multiple comparisons method
information and identifies the major difference between 2.1 The concept of multiple comparisons method
means of a set of group from the population under Multiple comparison methods are used to investigate
consideration [7]. differences between pairs of population means. In specific it
Whenever a researcher has more than two comparisons to test, investigates difference between subsets of population means
control of the Type I error rate becomes a concern. The using sample data. A set of confidence level is used to assess
analysis of variance (ANOVA) certainly helps the researcher the statistical difference between the means of groups. The
to identify existence of significant effect however it was confidence intervals allow to evaluate both statistical
created problem to control the type I error while dealing with significance and the practical significance of differences
multiple t-tests with multiple groups. Even if Fisher among means. Based on the basic characteristics of the
recommended using a more stringent alpha while performing hypothesis concept, the null hypothesis of no difference
his Least Significant Difference Procedure (LSD) but between means is rejected if zero is not contained in the
researchers finds the LSD process inadequate control of Type confidence interval [11].
I error [8, 9]. The problem associated with fisher method and
the hunger to get better statistical methods leads to 2.2 Types of multiple comparison method and its use
The selection of the appropriate multiple comparison method
development of several multiple comparison methods. Out of
depends on the desired inference. It is always not useful to use
these several developed methods, the most popular process
the Tukey all pair wise approach when other tests are suitable
for comparison of means between the pairs or groups in a
for study. Tukey confidence intervals are always wider and
population is the Tukey’s Honestly Significance Difference the hypothesis tests are less powerful for a particular family
Procedure (HSD) [10]. From this it can be clearly identified the error rate. Dunnett is preferable when the study has a control
role of selection of significance level (alpha value) is highly group and several treatment groups and the aim is to compare
crucial in multiple comparison test. Tukey's HSD is probably the treatment group with the control group. In this instant,
the most recommended and used procedure for controlling MCB is always preferable as compared to Dunnett if the
Type I error rate when making multiple pair wise requirement is to identify the groups that are the best. The
comparisons. The purpose of this study was to compare choice of Tukey versus Fisher's LSD methods mainly depends
observed Type I errors with nominal confidence level of 90 on the error rate i.e. based on requirement of error rate; it
%, 95 % and 99% in special education setting. might be family error rate individual error rate [12].
Table 1: Characteristics and advantages of different multiple comparison methods [11, 13, 14].
Normal Comparison with a Pair wise
Method Strength
Data Control Comparison
Tukey Yes Highly recommended test when doing all pair wise comparisons. No Yes
Dennett Yes Most influential test when comparison is with a control. Yes No
Hsu's MCB The most powerful test when comparison is done with the group with the
Yes No Yes
method highest or lowest mean to the other groups.
Games- Howell Yes Used when equal variances is not primary criteria No Yes

3. Tukey HSD Test process and came up with the test that bears his name. The
The Tukey HSD ("honestly significant difference" or "honest Tukey’s test compares the differences between means of
significant difference") test is a statistical tool used to values rather than comparing pairs of values. The value of the
determine if the relationship between two sets of data is Tukey test is given by taking the absolute value of the
statistically significant – that is, whether there's a strong difference between pairs of means and dividing it by the
chance that an observed numerical change in one value is standard error of the mean (SE) as determined by a one-way
causally related to an observed change in another value. In ANOVA test. The SE is in turn the square root of (variance
other words, the Tukey test is a way to test an experimental divided by sample size) [19].
hypothesis. The Tukey test is invoked if the interaction among The Tukey test is a post hoc test in that the comparisons
three or more variables is mutually statistically significant, between variables are made after the data has already been
which unfortunately is not simply a sum or product of the collected.
individual levels of significance [15, 16].
Tukey's method is used in ANOVA to create confidence 3.2 Steps wise process to perform Tukey HSD Test
intervals for all pair wise differences between factor level The Tukey HSD test analyzes all pair wise comparisons
means while controlling the family error rate to a specified among means. For measurement or analysis the HSD for each
level. It is essential to carefully evaluate the family error rate pair of mean was calculated by using the below formula [20]:
when making multiple comparisons because the chances of
making a type I error for a series of comparisons is greater 𝑀𝑖−𝑀𝑗
𝐻𝑆𝐷 =
than the error rate for any one comparison alone. To reduce √𝑀𝑆𝑤⁄𝑁

the higher error rate, Tukey's method calculate and adjusts the
confidence level for all individual comparisons that produces Where:
the family error rate so that the resulting simultaneous 1. Mi – Mj is the difference between the pair of means.
confidence level is equal to the specified value [17, 18]. (To calculate this, Mi should be larger than Mj)
2. MSw is the Mean Square Within, and N is the number in
3.1 Need of Tukey HSD Test the group or treatment.
Initially the results of multiple test studies conclusion were
drawn based on the alpha value and independent P value. Steps involved
John Tukey later realized the drawback associated with this Step 1: Perform the ANOVA test. Assuming your F value is
significant; run the post hoc test
~60~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

Step 2: Choose two means from the ANOVA output. Note the After the identification of the sample, the motor skills
following: (running, jumping and sliding) of individual subjects were
 Means, assessed by using the motor skill assessment scale. The
 Mean Square Within, observations were recorded on the observation sheet of the
 Number per treatment/group, record booklet and scoring was done in the assessment
 Degrees of freedom within. manual. The Numerical score and raw score of all domains
were calculated by using the assessment manual.
Step 3: Perform the Tukey’s HSD statistics by using the
above formula. Item No. Domains of motor skill development Assessment
Step 4: Determine the score reported in Tukey’s critical value Running/Jumping/Sliding
table. A. Not Responded (N)
Step 5: compare the values obtained in step 3 and step 4. If B. Occasionally Responded or try to respond (O)
the calculated value from Step 3 is larger than the value C. Responded well (R)
obtained in step 4 then it can be concluded that means are
significantly different. 5. Presentation of the case study for control of type I error
(with respect to significance level) while designing HSD
4. Case study for control of type I error (with respect to experiment
significance level) while designing HSD experiment using As per the study design the motor skill development scores
statistical tools after effective use of music for ten students in each group
A case study was presented for the process of performing were collected after implementation of music. The details of
HSD during investigation in a special education setting. The scores of the students with their mean, minimum and
study focuses cognitive effect of music on growth and maximum marks obtained are presented in the subsequent
enhancement of the motor behavior (running, jumping and section.
sliding) of children with mild intellectual disability enrolled in
the special school setting located in the Ganjam District of 5.1 Effect of type 1 error in HSD experiment consisting of
Odisha. The case study includes sample selection, design of 3 groups with different significance level (90, 95 and 99
study, tools used and effectively shown the affect of type I %)
error with change in significance level. The case study The effect of type I error in HSD experiment was performed
explains the experiment wise Type I error control with respect by taking three groups. The details of numerical scores of
to different significance level. gross motor skills obtain by the intellectual disability students
consisting of 3 groups (group 1, group 2 and group 3) with
4.1 Selection of Sample their mean, minimum and maximum scores obtained are
The population of the study will consist of 10 intellectual presented in Table 1.
disable special school children with mild mental retardation.
All intellectual disable students belong to mild category. The Table 1: Individual scores obtained by students (%) consisting of 3
groups (group 1, group 2 and group 3)
study will be conducted with the help of “BALVIKAS”
located near Kodola town of Ganjam district Odisha, INDIA. Individual scores obtained by students (%)
The study also considers only male student as an integral part Students consisting of 3 groups
of the study. These students were selected based on their Group 1 Group 2 Group 3
Intelligence Quotient (IQ) and between age group 8-15 years 1 82.64 65.83 48.21
[20]
. 2 66.69 75.12 44.26
3 62.03 54.22 65.42
4.2 Tools used 4 80.02 41.36 55.22
5 61.24 42.28 42.22
The tools used are mainly the available tools used by the
6 76.32 51.26 56.23
professionals in the field of statistics and intellectual
7 62.21 38.24 41.25
disability. The motor skill of individual subject was assessed 8 48.22 45.21 44.82
in terms of score (percentage) and appropriate tools will be 9 53.21 65.24 52.26
identified, developed and administered. The main tool used 10 44.45 56.84 41.26
for the evaluation of this study was statistical software Mean 63.703 53.56 49.115
Minitab 17.3.0. Min 44.45 38.24 41.25
Max 82.64 75.12 65.42
4.3 Scoring and interpretation
First, basic information (such as name, age, sex, IQ and level 5.2 HSD experiment consisting of 3 groups with different
of class) were collected for all the children with mild mental confidence level (90, 95 and 99 %) or error rate (1, 5, and
retardation enrolled in the study. Then the children were 10 %)
subjected to the matching procedures on the variables The one-way analysis of variance (ANOVA) is used to
mentioned (i.e. age, sex and IQ). Thus, a total of 10 determine whether the mean of a dependent variable is the
intellectual disable children with mild mentally retardation are same in two or more independent variable. ANOVA only
selected for each group. In the similar way 3 groups are helps to determine that at least two groups were different in
created and totally 30 intellectual disable children with mild the studied population. In the discussed case study one way
mentally retardation are enrolled. The age limit for the ANOVA was applied with significance level of 0.05 (α) to
children was 8 -15 years and the IQ considered was 40-50 for identify which group is different from all the three studied
group 1, 50-60 for group 2 and 60-70 for group 3 groups (three groups i.e., scores obtained by group 1, group 2
respectively. and group 3).

~61~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

Both one-way analysis of variance (ANOVA) and the Tukey other or not. The following method was adopted to perform
HSD test were performed simultaneously by using statistical the ANOVA study.
software Minitab 17.3.0. However, during Tukey HSD test
different error rate of comparison or confidence level (90, 95 6. Method
and 99 % Cl) were taken to compare the mean score of three Null hypothesis: All means are equal
groups. The interval plot of scores vs group (with 95 % CI of Alternative hypothesis: At least one mean is different
mean) from one way ANOVA and Tukey simultaneous with Significance level: α = 0.05
different significance level (90, 95 and 99 %)) were plotted. (Equal variances were assumed for the analysis)
The details of ANOVA summary and HSD test was presented
in Table 2 and 3 Table 2: Analysis of variance table
Adj Sum of
5.3 One-way ANOVA: Score versus Group Source DF Adj Mean square F-Value P-Value
square
The one way ANOVA test was performed to identify whether Group 2 1118 559.1
the specific groups were significantly different from each Error 27 3451 127.8 4.37 0.023
Total 29 4569 -

Fig 1: One way ANOVA plot of score vs group (3 groups) with 95 % CI for the mean

The result of one way ANOVA reveled that P value was also called post hoc tests, are the right tools to address this
found to be 0.023 which signifies that there is significant issue. Thus, Tukey pair wise comparisons test was selected
difference exist in the scores obtained in different groups. The and the difference in means of the group was determined.
one way ANOVA plot in Fig. 1 also signifies that the mean
values of all three groups interval distributed far from their 6.1 Tukey Pair wise Comparisons
respective means. As discussed earlier Tukey HSD test was performed with
From Table 1 and Fig 1, it was identified that ANOVA different error rate of comparison (1,5 and 10 %) or
detects a significant effect of scores between the groups. confidence level (90, 95 and 99 %) to compare the mean score
However, there is no information which specific population of three groups. The Tukey simultaneous with different error
groups differ significantly from one another. For this purpose, rate (90, 95 and 99 %) was presented in Table 3, Table 4 and
there is a requirement to test the differences between pairs of Fig 2.
groups. In this case the pair wise multiple comparisons tests,

Table 3: Grouping information (3 groups) using the Tukey Method with 90 %, 95 %, and 99 % confidence
Grouping information using the Tukey Method and 95% confidence
Group N Mean Grouping
Group 1 10 63.70 A
Group 2 10 53.56 AB
Group 3 10 49.11 B
Grouping information using the Tukey Method and 99% confidence
Group 1 10 63.70 A
Group 2 10 53.56 A
Group 3 10 49.11 A
Grouping information using the Tukey Method and 90% confidence
Group 1 10 63.70 A
Group 2 10 53.56 AB
Group 3 10 49.11 B
Note: Means that do not share a letter are significantly different.

The results of the Tukey's test are included in the grouping the above table specifically grouping information table helps
information table (Table 3), which highlights the significant to rapidly determine whether any significant difference
and non-significant comparisons. The information captured in between means of any pair of groups subsist or not. The
~62~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

grouping column presented with alphabetical letters that trend is also observed for in 90 % confidence interval or error
group the factor levels. Groups that do not share a letter have rate. However, in 99 % confidence it was observed that all the
a mean difference that is statistically significant. Table 3 three groups are significantly different. It is might be due to
shows that in 95 % confidence only group 2 shares both letter selection of 1 % error rate. The selection of 99 % confidence
A and B. However, it was observed that group 1 and 3 do not level or 1 % error rate makes the test more sensitive to
share a letter, which indicates that group 1 and 3 are detecting differences, thus a small change in means shows a
significantly different. It also provides evident that group 1 significant difference.
has a significant higher mean as compared to group 3. Same

Table 4: Tukey simultaneous tests for differences of means for 95 %, 90 % and 99 % confidence level
Tukey simultaneous tests for differences of means for 95 % confidence level
Difference of Levels Difference of means SE of difference 95 % CI T-value Adjusted P-value
Group 2 - Group 1 -10.14 5.06 (-22.69, 2.41) -2.01 0.130
Group 3 - Group 1 -14.59 5.06 (-27.14, -2.04) -2.89 0.020
Group 3 - Group 2 -4.44 5.06 (-16.99, 8.10) -0.88 0.658
Individual confidence level = 98.04%
Tukey simultaneous tests for differences of means for 90 % confidence level
Difference of Levels Difference of means SE of difference 90 % CI T-value Adjusted P-value
Group 2 - Group 1 -10.14 5.06 (-20.98, 0.69) -2.01 0.130
Group 3 - Group 1 -14.59 5.06 (-25.42, -3.76) -2.89 0.020
Group 3 - Group 2 -4.44 5.06 (-15.28, 6.39) -0.88 0.658
Individual confidence level = 95.87%
Tukey simultaneous tests for differences of means for 99 % confidence level
Difference of Levels Difference of means SE of difference 99 % CI T-value Adjusted P-value
Group 2 - Group 1 -10.14 5.06 (-26.23, 5.94) -2.01 0.130
Group 3 - Group 1 -14.59 5.06 (-30.68, 1.50) -2.89 0.020
Group 3 - Group 2 -4.44 5.06 (-20.53, 11.64) -0.88 0.658
Individual confidence level = 99.63%

(A)

(B)

~63~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

(C)
Fig 2: Tukey Simultaneous test for differences of mean of score for 3 groups with (A) 95% CI (B) 99% CI and (C) 90% CI

7. Observation and discussions 9. The conflict results obtained with 99 % confidence level
The confidence intervals in the Table 4 and Fig 2 for the for group 1-group 3 is might be due to improper selection
above case study indicate the following: of error rate which produces type I error highly
1. The 95% simultaneous confidence level indicates that we prominent in multiple comparisons test like Tukey’s test.
are 95% confident that all of these confidence intervals 10. The experiment-wise error rate represents the probability
contain the true differences. The same is true for 90% and of a type I error (false positive) over the total family of
99 % confident level. comparisons. The simultaneous confidence level is based
2. The 95 % confidence level for the difference between the on both the individual confidence level and the number of
means of group 1- group 2 and group 2 – group 3 extends confidence intervals. In 99 % confidence level the
from -22.69 to 2.41 and -16.99 to 8.10 respectively. This individual confidence level also found to be 99.63%. The
range include zero, which indicates that the difference observed type I error with 99 % simultaneous confidence
between these means is not statistically significant. level is resulted due to the high individual confidence
3. 95 % confidence level for the difference between the level.
means of group 1- group 3 extends from -27.14 to -2.04. 11. In post hoc tests like Tukey’s HSD, simultaneous
The confidence levels for this range does not include confidence level was used rather than an individual
zero, which indicates that the differences are statistically confidence level. The simultaneous confidence level
significant. applies to the entire family of comparisons. With a 99%
4. Each individual confidence interval has a confidence simultaneous confidence level, it can be 99% confident
level of 98.04%. This result indicates that 98.89% that all intervals in the set of comparisons contain the
confident that each individual interval contains the true actual population differences between groups. However,
difference between a specific pair of group means. The getting a 99 % confidence in an experiment is highly
individual confidence level for each comparison produces difficult. This makes the tests more sensitive to commit
the 95% simultaneous confidence level for all three error and this is the probable reason that leads to
comparisons. committing a type I error as compared to 90 % and 95 %
5. The 90 % confidence level for the difference between the confidence level.
means of group 1- group 2 and group 2 – group 3 extends
from -20.98 to 0.69 and -15.28 to 6.39 respectively. This 8. Recommendations
range include zero, which indicates that the difference In the above case study one-way analysis of variance
between these means is not statistically significant. (ANOVA) followed by Tukey HSD test was performed to
6. 90 % confidence level for the difference between the identify differences between the means of pairs of groups.
means of group 1- group 3 extends from -25.42 to -3.76. Three error rates are studied and committing of type I error
The confidence levels for this range does not include was evaluated for each error rate. From the result it was
zero, which indicates that the differences are statistically identified that with 90% and 95 % error rate or confidence
significant. level the probability of committing type I error is less.
7. The 99 % confidence level for the difference between the However, a type I error was observed with 99 % error rate.
means of group 1- group 2, group 1 group 3 and group 2 The possible reason of the observed type I error was identified
– group 3 include zero in their respective range, which and based the observed data and the results following
indicates that the difference between these means is not recommendations are proposed
statistically significant. 1. Error rate or confidence level should be taken carefully
8. However, from Table 4, it was observed that at 99% before commencing Tukey HSD test. Probability of
confidence level the adjusted P vale for group 1 – group 3 committing type I error increases with improper selection
was 0.020 which was lesser than selected significance of confidence level.
level. The P value indicated that there is significant 2. The size of population always plays an important role.
difference exist between group 1 – group 3. The same Sample size is also a significant predictor; as sample size
result also reflected in the grouping data. increases the difference decreases. Thus, it is always
~64~
International Journal of Statistics and Applied Mathematics http://www.mathsjournal.com

advisable to select larger number of populations in the 4. Bender R, Lange S. Adjusting for multiple testing-when
pair or group to minimize the type I error rate while and how. Journal of Clinical Epidemiology 2001;54:343-
performing Tukey HSD test. 349.
3. The probability of committing a type I error also depends 5. Benjamin JY, John HB. Tukey’s contributions to multiple
on the no of pairs taken to compare the mean. It is comparisons. The Analysis of Statistics 2002;30(6):1576-
assumed that when the number of groups increases the 1594.
difference between experiment wise errors proportions 6. Koch GG, Gansky SA. Statistical considerations for
also increases. Thus, it is also advisable to perform Tukey multiplicity in confirmatory protocols, Drug Information
HSD test with suitable numbers of pairs or groups to Journal 1996;30:523-534.
compare the mean. The selection of error rate or the 7. Sankoh AJ, Huqu MF, Dubin N. Some comments on
confidence level plays a significant role while performing frequently used multiple endpoint adjustment methods in
HSD test in pairs or groups. clinical trials. Statistical Method 1997;16(25):29-42.
8. Gill JL. Current status of multiple comparisons of means
9. Conclusion in designed experiments. Journal of dairy science
The selection of statistical parameters plays a significant role 1973;56(8):973-977.
for statistical analysis of any data with more than two 9. Jaccard J, Becker MA, Wood G. Pair wise multiple
comparisons. In fact, the search of appropriate conditions comparison procedures: a review. Psychological Bulletin
such as level of significance may be considered helpful for the 1984;96(5):89-96.
researcher or statistician to draw meaningful conclusion. 10. Toothaker LE. Multiple Comparison Procedures. SAGE
However, the research output also significantly depends on Publications 1993.
type I error. In statistical hypothesis testing, a type I error is 11. Minitab 18 Support: What is Tukey’s method for
the rejection of a true null hypothesis (also known as a "false multiple comparison Available at
positive") during drawing conclusion. Controlling appropriate https://support.minitab.com/en-us/minitab/18/help-and-
error rates is highly essential in order to protect against wrong how-to/modeling-statistics/anova/supporting-
conclusions. topics/multiple-comparisons/what-is-tukey-s-method.
In the present study the importance of selection of error rate Accessed on 02/12/2020.
or confidence level was explained with a study conducted 12. Ryan TA. Multiple comparisons in psychological
with the help of special school of BALVIKAS situated in research. Psychological Bulletin 1959;56:26-47.
Kodola, Ganjam, India with ten mild special intellectual 13. Savitz DA, Olshan AF. Describing data requires no
disable students in three groups. The study was conducted adjustment for multiple comparisons. American Journal
with the obtained raw data and sequential approach was of Epidemiology 1998;147:813-814.
performed by using Tukey's HSD test to find the suitable 14. Steven C, Sawyer F. Analysis of Variance: The
confidence level to minimize the type I error. The one way Fundamental Concepts. The Journal of Manual &
ANOVA study was conducted and Tukey's HSD test was Manipulative Therapy 2000, 17(2).
performed with different error rate. The error rate was 15. SPSS one-way ANOVA with post Hoc tutorials
calculated and then significant of different significance level Available at https://www.spss-tutorials.com/spss-one-
with respect to Tukey's HSD test was explained by using P way-anova-with-post-hoc-tests-example/.
value of ANOVA test and grouping information obtained 16. Sangseok L, Dong KL. What is the proper way to apply
from the Tukey test. The case study demonstrated that the the multiple comparison test. Korean Journal of
confident level controls the type I error in Tukey's HSD test. Anesthesiology 2018, 71(5).
In the presented case study the conflict results obtained with 17. Barnette JJ, McLean JE. Type I Error Of Four Pair wise
99 % confidence level for group 1-group 3 is resulted due to Mean Comparison Procedures Conducted As Protected
the high individual confidence level. Based on the result, And Unprotected Test. Journal of Modern Applied
recommendations are proposed to minimize type I error while Statistical Methods 2005;4(2):446-459.
performing Tukey’s simultaneous test. 18. Cabra HJI. Multiple Comparisons Procedures.
Circulation 2008;1(17):698-701.
10. Acknowledgements 19. Holm SA. Simple sequentially rejective multiple test
The authors express their sincere thanks to Mr. Prana Krushna procedure. Scandinavian Journal of Statistics 1979;6:65-
Das, Asst. professor, DPMIASSE&T, Kodola, Ganjam, 70.
INDIA and staffs of BALVIKAS, Kodola, Ganjam, INDIA 20. Mahapatra APK, Nanda A, Mohapatra BB, Padhy AK,
for providing necessary support to carry out the above work. Padhy I. Concept of outlier study: The management of
outlier handling with significance in Inclusive education
Competing interests setting. Asian research Journal of Mathematics
Authors have declared that no competing interests exist. 2020;16(10):7-25.

11. References
1. Rafter JA, Abell ML, Braselton JP. Multiple comparison
methods for Means. SIAM Review 2002;44(2):259-278.
2. Salkind NJ. Encyclopedia of research design. SAGE
Publications 2010.
3. Chen SY, Yi ZFX. A general introduction to adjustment
for multiple comparisons. Journal of Thoracic Disease
2017;9(6):1725-1729.

~65~

View publication stats