[Downloaded free from http://www.pnrjournal.com on Thursday, November 28, 2019, IP: 175.143.71.

217]

Technical Notes

How to select appropriate statistical test?

Selection of appropriate statistical test is very important for the ordinal data. For example, rating score/scale for the
analysis of research data. Use of wrong or inappropriate color, taste, smell, ease of application of products, etc.
statistical test is a common phenomenon observed in
articles published in biomedical journals. [1-4] Wrong Interval data
statistical tests can be seen in many conditions like use of Interval data has a meaningful order and also has the
paired test for unpaired data or use of parametric statistical quality that equal intervals between measurements
tests for the data which does not follow the normal represent equal changes in the quantity of whatever is
distribution or incompatibility of statistical tests with the being measured. But these types of data have no natural
type of data, etc.[5] Because of the availability of different zero. Example is Celsius scale of temperature. In the
types of statistical software, performing the statistical tests Celsius scale, there is no natural zero, so we cannot say
become easy, but selection of appropriate statistical test that 70°C is double than 35°C. In interval scale, zero point
is still a problem. can be choosen arbitraly. IQ Test is also interval data as
it has no natural zero.
Selection of appropriate statistical tests depends on the
following three things: Ratio data
• What kind of data we are dealing with? Ratio data has all the qualities of interval data (natural
• Whether our data follow the normal distribution or order, equal intervals) plus a natural zero point. This type
not? of data is observed to be used most frequently. [4] Example
• What is the aim of the study? of ration data is height, weight, length, etc. In this type
of data, it can be said meaningfully that 10 m of length is
double than 5 m. This ratio hold true regardless of which
WHAT KIND OF DATA WE ARE DEALING scale the object is being measured in (e.g., meters or yards).
WITH? Reason for this is the presence of natural zero.

Usually our research data fall in one out of the following

four types of data: nominal data, ordinal data, interval WHETHER OUR DATA FOLLOW THE NORMAL
data, and ratio data. DISTRIBUTION OR NOT?
Nominal data This is the second prerequisite for selection of appropriate
In these kinds of data, observations are given a particular statistical test. If you know the type of data (nominal,
name. Like a person is observed to be ‘male’ or ‘female’ ordinal, interval, and ratio) and distribution of data
or Name of drug as generic or brand, etc. Nominal data (normal distribution or not normal distribution), selection
cannot be measured or ordered but can be counted. These of statistical test will be very easy. There is no need to
types of data are considered as categorical data but the check distribution in the case of ordinal and nominal data.
order of the categories is meaningless. Data that consist Distribution should only be checked in the case of ratio
of two classes like male/female or dead/alive are called and interval data. If your data are following the normal
binomial data, and those that consist of more than two distribution, parametric statistical test should be used and
classes like tablet/capsule/syrup are known as multinomial nonparametric tests should only be used when normal
data. Data of these types are usually presented in the form distribution is not followed.
of contingency tables like 2 × 2 tables.
There are various methods for checking the normal
Ordinal data distribution, some of them are plotting histogram, plotting
Ordinal data is also a type of categorical data but in this, box and whisker plot, plotting Q-Q plot, measuring
categories are ordered logically. These data can be ranked skewness and kurtosis, using formal statistical test for
in order of magnitude. One can say definitely that one normality (Kolmogorov-Smirnov test, Shapiro-Wilk test,
measurement is equal to, less than, or greater than another. etc). Formal statistical tests like Kolmogorov-Smirnov and
Most of the scores and scales used in research fall under Shapiro-Wilk are used frequently to check the distribution

61
Journal of Pharmaceutical Negative Results | October 2010 | Vol 1 | Issue 2

62
Goal

Comparison of two Comparison of three or Measure Prediction

Comparison of groups more groups association
one group to a between two
Description of
hypothetical vaariables
one group
value
Unpaired groups Paired groups

Unmatched Matche From another From several measured

Groups measured or binomial variables
variable

O N R, I O N
R, I

Fisher's test R, I O N
Chi-square (chi-square R,I O N R,I N
or ND NN for large
ND NND Binomial test D samples)
Cochrane
Q NND
One- Wilcoxon test Unpaired t Mann-Whitney ND NND ND
sample t test test Multiple
test logistic
Simple linear regressions
Repeated Friedman test regression
measure
Technical Notes

s
ANOVA Nonparam
etric
R,I O N R, I O N regression

R, I O N McNemar's
test Contingency Simple logistic
ND NND coefficients regression
[Downloaded free from http://www.pnrjournal.com on Thursday, November 28, 2019, IP: 175.143.71.217]

Proportion N NND

N NND
Paired t R, I O N
Wilcoxon test Spearman
correlation Multiple linear
Mean, Median, Chi-square regressions
SD interquartile ND NND test or
range Multiple nonlinear

One-way Kruskal-Wallis test Pearson correlation

ANOVA

R, I = Ratio and Interval data O= Ordinal data N = Nominal data

N = Normal distribution NND = Non normal distribution

Figure 1: Flow chart for selection of appropriate statistical tests

Journal of Pharmaceutical Negative Results | October 2010 | Vol 1 | Issue 2

[Downloaded free from http://www.pnrjournal.com on Thursday, November 28, 2019, IP: 175.143.71.217]

Technical Notes

of data. All these tests are based on null hypothesis that What kind of data we are dealing with?
data are taken from the population which follows the In the present study, the data type is nominal data.
normal distribution. P value is determined to see the alpha
error. If P value is less than 0.05, data is not following the Whether our data follow the normal distribution or not?
normal distribution and nonparametric test should be used There is no need to check distribution in nominal data.
in that kind of data. If the sample size is less, chances of They follow chi-square distribution.
non-normal distribution are increased.
So by looking at the table, we can say that most appropriate
test in this condition will be ‘Fishers Test’ (Chi-Square if
WHAT IS THE AIM OF THE STUDY? large sample size).

This is the third prerequisite of selection of appropriate Jaykaran

statistical test. What we want to compare? Whether Department of Pharmacology, Government Medical College,
we want to compare the drug with placebo? Or we Surat, Gujarat, India
want to compare effect of intervention by comparing Address for correspondence:
preinter vention endpoints with postinter vention Dr. Jaykaran, Department of Pharmacology,
endpoints. Government Medical College, Surat, Gujarat, India.
E-mail: drjaykaran@yahoo.co.in

If a researcher is clear about all the three questions

mentioned in previous text, appropriate statistical test can REFERENCES
be selected from the flowchart [Figure 1]; this flow chart
1. Lang T. Twenty statistical errors even you can find in biomedical
is modified from the table given on a site of software research articles. Croat Med J 2004;45:361-70.
graphpad (www.graphpad.com). 2. Jaykaran, Yadav P, Chavda N, Kantharia ND. Some issues related
to the reporting of statistics in clinical trials published in Indian
medical journals: A survey. Int J Pharmacol 2010;6:354-9.
To understand it better, an example has been discussed 3. Karan J, Kantharia ND, Yadav P, Bhardwaj P. Reporting statistics in
below. clinical trials published in Indian journals: A survey. Pak J Med Sci
2010;26:212-6.
4. Karan J, Goyal JP, Bhardwaj P, Yadav P. Statistical reporting in Indian
Incidence of tumors observed in control and treated mice Pediatrics. Indian Pediatr 2009;46:811-2.
in a preclinical study is as follows: 5. Strasak AM, Zaman Q, Karl PP, Gobel G, Ulmer H. Statistical errors
• Controls - 1 of 14 animals in medical research-a review of common pitfalls. Swiss MED Wkly
2007;137:44-9.
• Treated - 6 of 14 animals
Access this article online
Quick Response Code:
Use appropriate statistical test to determine if the incidence Website:
is significantly different in two groups.
www.pnrjournal.com

What is the aim of the study? DOI:

Our aim is to compare incidence of tumor between two
unpaired groups (control vs treated). 10.4103/0976-9234.75708

Staying in touch with the journal

1) Table of Contents (TOC) email alert
Receive an email alert containing the TOC when a new complete issue of the journal is made available online. To register for TOC alerts go to
www.pnrjournal.com/signup.asp.

2) RSS feeds
Really Simple Syndication (RSS) helps you to get alerts on new publication right on your desktop without going to the journal’s website.
You need a software (e.g. RSSReader, Feed Demon, FeedReader, My Yahoo!, NewsGator and NewzCrawler) to get advantage of this tool.
RSS feeds can also be read through FireFox or Microsoft Outlook 2007. Once any of these small (and mostly free) software is installed, add
www.pnrjournal.com/rssfeed.asp as one of the feeds.

63
Journal of Pharmaceutical Negative Results | October 2010 | Vol 1 | Issue 2