A new method for computing ability estimates when the response vector contains only errors or correct responses is proposed. It is based on the Stepsize method (Dodd, 1990). Instead of taking as the new ability the midpoint between the last computed ability and the corresponding extreme b- value, the new method proposes as the new estimate the mean of a truncated normal (0, 1) distribution. Last ability estimate and the correponding extreme b-value are the points in which the normal distribution truncates.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados