Discriminative Ability

In accessibility research methodology, the property of an evaluation metric to reveal statistically significant differences between stimuli that are known to differ along the dimension being measured. For example, a comprehension-question metric has discriminative ability for text complexity if it produces significantly different scores between texts engineered to be at different complexity levels. A metric without discriminative ability cannot be used to judge whether an intervention (such as Automatic Text Simplification) has changed the target property, even if the intervention did in fact change it. Researchers commonly pair discriminative ability with analysis of literacy bias (whether the metric systematically favours one participant sub-group over another).