Two One-sided Test

TOST requires two directional tests because an acceptable treatment difference must satisfy both sides of the clinical range. Rejecting only the hypothesis associated with the upper boundary, for example, would not establish that the difference also exceeds the lower boundary. The two-test requirement prevents a one-sided result from being interpreted as full equivalence.

Equivalence margins translate clinical acceptability into statistical hypotheses. The lower margin identifies one boundary of the acceptable difference, while the upper margin identifies the other. Because these limits are prespecified, the analysis evaluates treatment comparability against a clinically meaningful interval rather than against zero alone, making the result relevant to decisions about acceptable clinical differences.

The confidence interval provides an interval-based view of the possible treatment difference. Equivalence is supported when the entire interval lies within the prespecified lower and upper margins, because every value represented by that interval remains clinically acceptable. If any portion extends beyond a margin, the evidence does not satisfy the interval criterion for equivalence.

A test against zero asks whether the treatments differ, but it does not determine whether the difference is acceptably small. TOST addresses that clinical question by judging the observed difference against prespecified equivalence boundaries. This distinction is important when a new intervention may be considered comparable even though its measured effect is not exactly identical to the comparator.

The analysis begins by prespecifying the lower and upper clinical equivalence margins and selecting the significance level. Researchers then conduct the two directional tests, one for each boundary, and assess the corresponding confidence interval. Equivalence is concluded only when both one-sided null hypotheses are rejected and the interval remains contained within the specified margins.

TOST supports equivalence trials, comparisons of treatments, and assessment of new interventions when small differences are clinically acceptable. Its value lies in separating statistical difference from practical comparability: a result can be evaluated according to whether it stays within a clinically justified range. This makes the method relevant when exact equality is not required for clinical use.