# Sample size calculation: two ways of approaching it

Those involved in sample size calculation are typically either (bio)statisticians or epidemiologists.

Each has a different approach to sample size calculation. In this post we will look at the basic differences between them.

At the outset, I would like to point out that the choice of approach is dependent upon individual preference- one is not superior to the other.

The Biostatistician‘s approach:

The things considered are

1. Whether the variable is continuous or not

2. Which family of tests does it fall under (t tests, F tests, etc.)

3. Is it one-sided or two-sided?

4. What is the level of significance (alpha)?

5. What is the power (1-beta)?

6. What is the magnitude of effect?

7. How many groups are there? etc.

The Epidemiologist‘s approach:

The things considered are

1. What is the study design (case control, cohort, cross-sectional, RCT)?

2. How many groups are there (usually two)?

3. What is the ratio of the groups to each other (1:1; 1:2; 1:3, etc.)?

4. What is the desired/ expected Odds Ratio/ Risk Ratio (0.5, 1.5, 2, 3, etc)?

5. What is the proportion among the exposed?

6. What is the proportion among the unexposed?.

7. Is there matching? etc.

It is easy to identify the differences between the two approaches:

While the epidemiologists are more concerned with the study design and measures of risk/ association, the biostatisticians are more concerned about the type of variable and the family of tests. Typically, epidemiologists calculate sample size for a given measure of association (Odds Ratio of 2, for example). It is possible for biostatisticans to calculate sample size without having to consider the study design, rather basing the calculation upon the choice of statistical test that will be used during analysis (family of tests mentioned above).

What is the real-world impact of this difference?

If you approach a biostatistican to help with sample size calculation, you will be questioned about the type of variable (primary objective), effect size, and statistical tests for analysis.

However, if you approach an epidemiologist, you will be asked to specify the exact study design, proportions among exposed and unexposed, and the magnitude of Odds Ratio/ Risk Ratio.

Understandably, if you first consult a biostatistician, then an epidemiologist (or vice-versa), you are likely to be very confused. However, on the bright side, doing so will help you obtain clarity on not only the study design, but also the statistical analyses required (both requirements of a study protocol).

## 4 thoughts on “Sample size calculation: two ways of approaching it”

1. Anonymous

Nice Analysis…

Like

1. drroopesh Post author

Dear Someone,

Thanks. I hope you found it useful.

Best,
Roopesh

Like

2. ravi

sir
is it possible to get different sample size by using different approach

Like

1. drroopesh Post author

Dear Ravi,

Yes, one can obtain different estimates by using different approaches. Even within a single approach, merely choosing a different study design/ family of tests may result in a different value.
This is why the standard practice in the west is to calculate a range of values instead of a single estimate.

Regards,
Roopesh

Like

This site uses Akismet to reduce spam. Learn how your comment data is processed.