Recallab › Question bank › Social & Preventive Medicine › Biostatistics › #50141
Choosing a statistical test
An investigator measures haemoglobin in 60 women receiving iron supplementation and in 60 different women receiving no supplement, and wishes to compare the mean values. Haemoglobin is normally distributed in both groups. Which test is most appropriate?
- A. One-way analysis of variance
- B. Paired t-test
- correctC. Unpaired (independent samples) t-test
- D. Chi-square test
Explanation
Key fact
Two independent groups, one continuous normally distributed outcome, means an unpaired t-test. Paired is for the same people measured twice.
Unpaired t-test. The two groups are different women, so the observations are independent.
How it happens
- First ask what type the outcome is. Haemoglobin is continuous.
- Then ask whether it is normally distributed. Here it is, so a parametric test is allowed.
- Then ask how many groups. Two.
- Then ask whether they are paired. They are different women, so independent.
- Continuous, normal, two independent groups gives the unpaired t-test.
A short decision table
| Outcome | Groups | Test |
|---|---|---|
| Continuous, normal | Two independent | Unpaired t-test |
| Continuous, normal | Same subjects twice | Paired t-test |
| Continuous, normal | Three or more | Analysis of variance |
| Continuous, not normal | Two independent | Mann-Whitney U |
| Categorical | Two or more | Chi-square |
Three questions choose almost every basic test
Why not the others
- A. One-way analysis of variance — compares means across three or more groups. With two groups it reduces to the t-test anyway.
- B. Paired t-test — is for two measurements on the same individuals - before and after supplementation in one group, for instance. These are different women.
- D. Chi-square test — compares proportions in categorical data. It would be right for comparing the proportion anaemic, not the mean haemoglobin.
Remember
Type of outcome, number of groups, paired or not. Three questions choose almost every basic test.
Go deeper
The t-test assumes normality and roughly equal variances; with skewed data or small samples use the Mann-Whitney U test instead.
The Student t-test is parametric. The non-parametric tests are Mann-Whitney, Wilcoxon signed rank, Kruskal-Wallis and the sign test - a distinction the recalls ask directly.
Park K. Park's Textbook of Preventive and Social Medicine, 28th ed. (Bhanot, 2025).