Test equating with evolving latent ability

In large-scale assessments, students' ability is usually evaluated using multiple test forms, which require the use of several items. In this context, calibrating items before the official tests can be difficult for different reasons. A solution is to calibrate items during the first test administration and then use these estimates in the subsequent ones. However, this approach does not consider that the populations could be significantly different in terms of average ability, which is particularly problematic when the final output of this process is a merit ranking. Our findings show that, on one side, calibrating item parameters on populations with differences in ability does not affect the final merit ranking and, on the other side, the differences in item parameter estimates are significant.