MISSOURI UNIV-COLUMBIA TAILORED TESTING RESEARCH LAB
While numerous articles have appeared in the literature which describe the one-parameter logistic model and its application in a tailored testing setting, little or no research was conducted on the operational characteristics of the procedure when program parameters and item pool attributes are varied. The primary objective of this investigation was to determine the effects of varying the program parameters, stepsize and acceptance range, as well as the item pool attributes, size and shape, on the bias and standard error of the maximum likelihood ability estimates obtained from tailored tests. Specifically, two main research questions were addressed. First, what values of stepsize and acceptance range provided the last bias and smallest standard error of ability estimates. Secondly, what shape and size of item difficulty distribution provided the least bias and standard error of ability estimates across the range of the latent trait. In addition to the recommendations regarding the research questions stated above, an effort was made to discuss the interaction of the variables of stepsize, acceptance range, item pool size and the shape of the distribution of item pool difficulties. Results suggested that each of these variables played a substantial role in affecting the magnitude of statistical bias and standard error at various points along the ability continuum. The intent was to provide figures and tables to facilitate applications of tailored testing procedures such that a minimum of bias and standard error of ability estimates could be attained.