Routing in the Multistage End of Primary School Test
Maaike M. van Groen, J. Hendrik Straat, & Marie-Anne Keizer-Mittelhaëuser
Session 1A, 10:30 - 12:00, HAGEN 2
Major changes are currently made on the Dutch end-of-primary-school placement test. The main change consists of making new multistage tests available. Per 2018, a domain-specific multistage test is constructed following a 1-3-3 test design. All students will make the same first stage module. After this initial module, a routing decision is made. One of three modules is selected based on the student’s previous responses. After finishing the second module, a similar routing decision will be made. This implies that routing will take place twice for each of the three subjects. Given that six routing decisions will be made per student and the influence of routing on test results, it is important to consider routing carefully.
Many routing methods have been described in the literature. Several of these methods were investigated for this specific test using simulations. One challenging aspect about routing is that some methods also need prespecified cutoff points. For example, when raw scores are used as routing cutoff points one needs to determine their precise value. The student’s raw score is then compared with the cutoff points to determine the next module for the student. Although the routing method itself is simple and easy to compute, the method for specifying the cutoffs is more complicated. This specification can be done in a number of ways. One option is to specify the cutoffs such that equal proportions of students are routed through each path. Another option is to use simulations to determine the optimal cutoff points. Optimality is then determined using a criterion such as the precision of the reported ability estimates, the proportion of students per test path, or the classification accuracy. This implies that depending on the routing method, a second method can be required for specifying the cutoff points.
Depending on the routing method choices can be made regarding the input for the routing method. Are decisions based on all previously administered modules or on the last administered modules? Is maximum information computed for the next module or all remaining paths? Are all possible paths admissible? We will discuss the considerations for different choices and use simulations to demonstrate the effect of different choices.
Before the first test administration simulations were run to investigate the many choices that could be made for routing. Decision making about routing was supported by those simulations. After the first test administration in April 2018, the effects of those decisions will be evaluated. How many students took each path through the test? Which changes should be made in the routing procedure for the 2019 test administration? How can we further improve routing? These and other questions will be reflected upon based on the data from the 2018 test administration.