Accurate response times are essential for assembling tests in educational largescale assessment to ensure test validity and efficient testing. Because obtaining empirical response times in pilot studies is cost-intensive and also because this process is complicated in paper-and-pencil assessments, we propose a model-based approach for calculating response times from readily available testlet properties. This prediction formula was developed using the response time data of 334 high school students who worked on 93 testlets of a paper-and-pencil test measuring science achievement. A large proportion (94.3 %) of the variance in response times (i.e., the variation of the average response times of persons across testlets) was explained by number of items, number of words, and response type. Another sample of 1,386 students who worked on 125 additional science testlets was used to validate the initial findings. Overall, the proposed easy-to-use formula is suitable for providing accurate response times for test assembly at a low cost.

Keywords
Response time; Test assembly; Large-scale assessment