SN2 SHORT NOTE 5M ★ High-yield

Describe the Student t test and its applications.

Community Medicine · 3rd Professional

The Student T Test

The student t test is a parametric test used to compare means when the sample is small and the population standard deviation is unknown. It was devised by W. S. GOSSET, who published under the pen name "Student" because his employer, the Guinness brewery, forbade publication.

Why IT exists — with large samples the sample SD is a good estimate of the population SD and the normal (Z) distribution may be used; with small samples the sample SD is unreliable, and the t distribution allows for that extra uncertainty
The t distribution resembles the normal curve but is flatter with heavier tails, which means a larger value is required to reach significance; and it approaches the normal distribution as the sample size increases, becoming practically identical above about 30
The assumptions — the data are quantitative; drawn from a normally distributed population; the samples are random; and, for the unpaired test, the variances of the two groups are approximately equal
The three forms, and choosing correctly between them is the whole practical question:
1. The one-sample t test — compares a sample mean with a known or hypothesised population mean; for example, whether the mean haemoglobin of a group of pregnant women differs from a reference value
2. The unpaired (independent, two-sample) t test — compares the means of two independent groups; for example, mean birth weight in infants of smokers and non-smokers
Degrees of freedom = n1 + n2 - 2
3. The paired t test — compares two measurements on the same subjects, or on matched pairs; for example, blood pressure before and after treatment
It works by calculating the difference for each pair and testing whether the mean of those differences differs from zero, with degrees of freedom = n - 1, where n is the number of pairs
The choice between paired and unpaired IS the point most often got wrong, and the consequence is real
Data are paired where each observation has a natural partner — the same individual measured twice, twins, matched cases and controls, or two eyes or two limbs of the same patient
The advantage of pairing is that each subject acts as their own control, which removes between-individual variability and therefore gives greater power with fewer subjects
Applying an unpaired test to paired data throws away that advantage and may miss a real effect; applying a paired test to independent data is simply invalid
Interpretation — the calculated t is compared with the tabulated value for the appropriate degrees of freedom; if the calculated value exceeds the tabulated one, the null hypothesis is rejected
Always report the mean difference with its confidence interval alongside the p value, since the p value alone conveys nothing about the size of the effect
When the assumptions fail — use the MANN-whitney U test in place of the unpaired t test, and the wilcoxon signed-rank test in place of the paired t test; or transform the data, commonly by taking logarithms
For more than two groups the t test must not be used repeatedly, since each comparison carries its own 5% error and the overall error rate inflates; ANOVA is the correct test
Paired data need a paired test, and using the wrong one throws away real statistical power
Each subject acts as their own control when paired, which removes between-person variability
Two eyes of one patient are paired, as are twins and matched case-control pairs
Never run repeated t tests across groups; each comparison carries its own five per cent error
The t distribution has heavier tails, so a larger value is needed to reach significance in small samples
Above about thirty it becomes the normal curve, which is why the distinction matters only for small samples
The paired test works on the differences, asking whether their mean differs from zero
Degrees of freedom differ between the forms, being n minus one for paired and n1 plus n2 minus two for unpaired
Report the mean difference with its interval, since a p value alone says nothing about the size of the effect
Mann-Whitney replaces the unpaired t test and Wilcoxon the paired one, where normality fails
Log transformation may restore normality, and is often preferable to abandoning a parametric test
Equal variances are assumed in the unpaired test, and should be checked rather than presumed
Gosset published as Student because his employer forbade it, which is the origin of the name

CLINICAL PEARL

Paired data need a paired test, and getting this wrong costs real power. The same person measured twice, twins, matched pairs and two eyes of one patient are all paired — and the advantage of pairing is that each subject acts as their own control, removing between-person variability. And never run repeated t tests across several groups: each comparison carries its own 5% error, which is what ANOVA exists to prevent.

All questions in Biostatistics

Every Community Medicine chapter →