Generally Applicable Variance Estimation Methods for Common-Population Linking

Educational assessments require periodic administration changes, such as transitioning from paper to digital administration. During such transitions, a linking function relating the metrics of the new and previous administration is estimated, but uncertainty in this estimation introduces variance into comparisons between administrations. Similarly, different assessments provided to overlapping populations may be linked, introducing linking or equating error. We introduce new generally applicable variance estimation methods, generalize prior methods to be more widely applicable, and confirm the validity of the methods via simulation. Our methods account for dependencies between linking and other sources of error, complex sampling, and nonlinear linking functions, while applying to a wide range of score comparisons and statistics such as means, standard deviations, percentiles, and differences.