The Effects of Different Parallelism Assumptions on the Estimates of Classification Consistency Using IRT and non-IRT Approaches