Unproctored Internet Test Verification
针对高利害无人监考网络测试，提出两种方法验证原始测试与后续确认测试的作答一致性，模拟表明基于序贯概率比检验的自适应确认测试在保持检测效力同时长度缩短近四倍。
Unproctored Internet testing (UIT) is commonly used in employment test administration. When the test is high stakes, the International Guidelines on Computer-Based and Internet-Delivered Testing recommend to follow up the results with a confirmation test in a controlled setting. This article proposes and compares two methods for detecting whether a test taker’s original UIT responses are consistent with the responses from a follow-up confirmation test. The first method is a fixed length adaptive confirmation test using the likelihood ratio (LR) test to evaluate cheating and the second method is a variable length adaptive confirmation test using an extension of the stochastic curtailed truncated sequential probability ratio test (SCTSPRT) to evaluate cheating. Simulation studies indicated that the adaptive confirmation test using the SCTSPRT was almost four times shorter while maintaining the same detection power. The study also demonstrated that cheating can have a detrimental effect on the validity of a selection procedure and illustrated that the use of a confirmation test can remedy the negative effect of cheating on validity.