The Rasch model for IRT only indicates the position of each item on the latent trait. An additional parameter is required to indicate how sharply an item demarcates people above and below that level—a notion closely related to sensitivity and specificity (61). The two-parameter model adds information on item discrimination, plotted on a vertical axis that indicates the cumulative probability of endorsing each item as the level of the trait increases. This produces Sshaped, normal ogive “item characteristic curves” for each item starting at the base of the graph and sloping up more or less steeply to meet the top line; a curve is produced for each item. As before, the distance of each curve from the left of the graph indicates the threshold or severity of the trait at which the item will be answered positively; the slope of the curve indicates the discriminal ability or accuracy of the item. The ideal is a steep slope, suggesting that the item sharply demarcates people along the trait. In practice, item slopes may vary across the severity range, with severe symptoms often forming more reliable indicators than mild symptoms: it is harder to write a good questionnaire item reflecting mild illness than severe (58, p401).