19
Review Article | Sample size determination
the dierence between means of the weight
reduction (which constitutes part of the eect
size for independent sample t-test) should be
suciently large to demonstrate the superiority
of the new diet programme over the conventional
diet programme.
In the second category, the research
rationale is to measure accurately the
eectiveness of the new diet programme to
reduce weight in comparison with conventional
diet programme, irrespective of whether the
dierence between both programmes is large or
small. In this situation, the dierence does not
matter since the researcher aims to measure an
exact dierence between them, which means
that it can only tolerate a very low margin of
dierence. In this circumstance, the researcher
will therefore only be able to accept the smaller
eect sizes. The estimate of eect sizes in this
instance can be reviewed either from literatures,
pilot study, historical data and rarely by using an
educated guess.
The acceptable or desirable eect size that
can be found from the literature can vary over a
wide range. Thus, one of the better options is to
seek for the relevant information from published
articles of recent studies (within 5 years) that
applied almost similar research design such
as used the same treatments and had reported
about similar patient characteristics. If none
of these published articles can provide a rough
estimate of the desired eect size, then the
researcher may have to consider conducting
a pilot study to obtain a rough estimate of the
closest approximation to the actual desired eect
size. Besides, historical data or secondary data
can also be used to estimate the desired eect
size, provided that the researcher has access to
the secondary data of the two diet programmes.
However, it must be emphasised that deriving
the eect size from secondary data may not
always be feasible since the performance of the
new intervention may still not yet have been
assessed.
The last option is to estimate the desired
eect size based on a scientically or a clinically
meaningful eect. This means the researcher,
through his or her own knowledge and
experience, is able to determine an expectation
of the dierence in eect, and then to set a target
dierence (namely, eect size) to be achieved.
For example, a researcher makes an educated
guess about the new diet programme, and
requires it to achieve a minimum dierence of
3 kg in weight reduction per month in order for it
Step 2: To Select the Appropriate Statistical
Analysis
Researchers have to decide the appropriate
analysis or statistical test to be used to answer
the study objective; regardless of whether
the aim is to determine a single mean, or a
prevalence, or correlation, or association, just
to name a few. The formula that will be used to
estimate or calculate the sample size will be the
same as the formula for performing the statistical
test that will be used to answer the objective of
study. For example, if an independent sample
t-test has to be used for analysis, then its sample
size formula should be based on an independent
sample t-test. Hence, there is no a single formula
for sample size calculation or estimation which
can apply universally to all situations and
circumstances.
Step 3: To Calculate or Estimate the Sample Size
Estimating or calculating the sample
size can be done either by using manual
calculation, sample size software, sample size
tables from scientic published articles, or by
adopting various acceptable rule-of-thumbs.
Since both the type I and type II errors are
already pre-specied and xed, hence only the
eect size remains to be specied in order for
the determination of an appropriate sample
size. To illustrate this point, it will be easier
to demonstrate by using a case scenario as an
example. Say a researcher would like to study an
eectiveness of a new diet programme to reduce
weight. The researcher believes the new diet
programme is better than the conventional diet
programme. It was found that the conventional
diet programme can reduce on average 1 kg in
1 month. How many subjects are required to
prove that the new diet programme is better than
the conventional diet programme?
Based on Step 1 and Step 2, a researcher
has decided to apply the independent sample
t-test to answer the objective of study. Next, the
researcher will need to specify the eect size after
having both type I error and power set at 0.05%
and 80%, respectively (type II error = 20%).
What margin of eect size will be appropriate?
This shall depend on the condition itself or the
underlying research rationale which can then
be further classied into two categories. In the
rst category, the research rationale is to prove
that the new diet programme (for reducing
weight) is superior to the conventional diet
programme. In this case, the researcher should
aim for sizeably large eect size. In other words,