Selecting Statistical Tools

SELECTING
STATISTICAL TESTS
~ Lovelyn B. Lasac – Chen
What determines the use of statistics?
Types of Scale
Levels of Measurement
Statistics
THE QUESTIONS
These will guide you toward the appropriate
statistical test.
How does my sample compare to the general population?
These types of question explore a particular

variable in relation to a population. For
example, say you are conducting a study of
university students. You might ask if the
percentage of males/females in your sample,
or their average age, or even their opinions,
approximate that of the broader student
population. Of course to answer these
questions you will need population data for
the greater student body.
Are there differences between two or more
groups of respondents?
These types of question are very common

and can be referred to as ‘between subject’.
For example, you might ask if male and
female university students are likely to be
different majors; or whether students of
different ethnic backgrounds are likely to
rank subjects differently; or whether
students graduating with different majors
have different starting salaries.
Have my respondents changed over time?
These types of question involve before and

after data with either the same group of
respondents or respondents who are
matched by similar characteristics. They are
often referred to as ‘within subject’. An
example of this type of question might be
whether students’ perceptions of the
importance of university changes after their
first semester.
Is there a relationship between two or more
variables?
These types of question can look for either
correlations (simply an association) or cause
and effect. Examples of correlation question
might be, ‘Is there an association between
hours of study and perceived value of study?’
or, ‘Is there a correlation between how much
you enjoy university and the type of job you
are likely to get upon graduation?’
Is there a relationship between two or more
variables?
Questions looking for cause and effect
differentiate dependent and independent
variables. For example, ‘Does starting salary
depend on major?’ Or, ‘Does course
satisfaction depend on hours of study?’
Cause-and-effect relationships can also look
to more than one independent variable to
explain variation in the dependent variable.
For example, ‘Does starting salary depend on
a combination of major, age, and socio-
economic status?’
‘Not everything that can be counted counts,
and not everything that counts can be counted.’
–Albert Einstein
Two considerable concerns:
(1) there are many social meanings that
cannot be reduced to numbers; and
(2) even if this reduction were possible, there
would be a loss of ‘richness’ associated with
the process.
Statistics and number of groups
Statistical Analysis
Selecting Statistical Test
HOW TO CHOOSE THE RIGHT
STATISTICAL TEST?
1) Is there a difference between groups that
are unpaired?
• Different tests are required for quantitative

or numerical data and qualitative or
categorical data as shown in the figure.
• For numerical data, parametric tests are
applied if they follow the parameters of the
normal distribution curve (Gaussian curve).
• Non-parametric tests are used if

distribution of the data is not normal or if
one is not sure about the distribution.
When comparing more than two sets of
numerical data, a multiple group comparison
tests should be used first such as:
• One-way analysis of variance (ANOVA)

• Kruskal-Wallis test
P value Post hoc
(usually meaning p < 0.05) test
• Repeatedly applying the t test or its non-

parametric counterpart to a multiple group
situation increases the possibility of
incorrectly rejecting the null hypothesis.
2) Is there a difference between groups
which are paired?
• Pairing signifies that data sets are derived

by repeated measurements.
• Pairing will also occur if subject groups are

different but values in one group are in
some way linked to values in the other
group (Twin Studies, Sibling Studies)
• Crossover study design also calls for the
application of paired group tests
• Sometimes subjects are deliberately paired

to match baseline characteristics such as:
- age
- sex
- severity or duration of disease
3) Is there any association between
variables?
• Various tests applicable are outlined in the

figure.
4) Is there agreement between data sets?
• Agreement between numerical variables

may be expressed quantitatively by the
intraclass correlation coefficient or by
Bland-Altman plot.
• In case of categorical data, the Cohen’s

Kappa statistic is frequently used.
5) Is there a difference between time-to-event
trends of survival plots?
• If there are two groups, then the
applicable tests are Cox-Mantel test,
Gehan’s (generalized Wilcoxon) test of
log-rank test.
• In case of more than two groups, Peto

and Peto’s test of log rank test can be
applied.
Example 1.

Selecting Statistical Tools

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Selecting Statistical Tools

Uploaded by

Copyright:

Available Formats

SELECTING

These types of question explore a particular

These types of question are very common

These types of question involve before and

• Different tests are required for quantitative

• Non-parametric tests are used if

• One-way analysis of variance (ANOVA)

• Repeatedly applying the t test or its non-

• Pairing signifies that data sets are derived

• Pairing will also occur if subject groups are

• Sometimes subjects are deliberately paired

• Various tests applicable are outlined in the

• Agreement between numerical variables

• In case of categorical data, the Cohen’s

• In case of more than two groups, Peto

You might also like