测验的目标是得到被试潜在特质的有效估计值。在现代测量理论中（例如，项目反应理论），很多测量模型的建立都基于被试在整个测验过程中保持努力作答状态这一假设。然而，很多研究发现，不努力作答行为在教育测验中非常常见。当测验作答反应中含有不努力作答时，如果直接使用传统测量模型对数据拟合，往往会得到有偏差的估计结果。为了尽可能消除不努力作答的不利影响，应对原始数据中的不努力作答进行识别和清理。

近年来，随着计算机测试的普及，研究者利用在测验中获得的反应时等信息，提出了各种方法识别不努力作答，以提高测量模型参数估计结果的准确性。本研究基于传统标准化反应时残差法，提出了识别不努力作答的新方法——固定参数迭代标准化残差法。其基本思想是，第一步识别努力作答被试群体并估计得到题目参数，第二步固定题目参数并基于标准化残差法进行迭代净化，识别不努力作答。本研究采用模拟研究方法系统地比较了不同条件下该方法与传统的标准化残差法相比的优势。此外，还在固定参数迭代标准化残差法体系下构建了用于评估和选择方法的判别指标，并探索了临界值。本研究一共包括三个模拟研究和一个实证研究，主要研究过程及其成果如下。

第一，证明了基于混合模型的方法能够识别努力作答被试群体，并基于此群体，得到较准确的题目参数估计结果。该方法同时利用了作答反应和反应时的信息，并且不包含强假设。研究一采用模拟研究方法，分别在速度为单组正态分布和混合正态分布的情境下，以样本量、题目数、不努力作答规模、不努力作答严重性、两种作答反应时差异为模拟条件，考察并比较了基于混合模型的方法和两种常用方法（基于混合反应时模型的方法、结合了反应时努力程度（RTE）指标的常模阈值法）在识别不努力作答被试和提高题目参数估计准确性方面的表现。结果表明，当不努力作答严重性较高或被试速度服从混合正态分布时，基于混合模型的方法具有较大优势。使用该方法筛选被试拟合模型，能够得到较准确的题目参数估计结果。

第二，提出了固定参数迭代标准化残差法，并探索了该方法识别不努力作答的有效性和适用条件。固定参数迭代标准化残差法将研究一中混合模型方法估计得到的题目参数估计值固定，在标准化残差法中加入了迭代净化的过程。研究二采用模拟研究方法，在样本量、题目数、不努力作答规模、不努力作答严重性、两种作答反应时差异不同的条件下，考察并比较了固定参数迭代标准化残差法、固定参数标准化残差法和原始标准化残差法在识别不努力作答和提高参数估计结果准确性方面的表现。结果表明，不努力作答规模越大，不努力作答严重性越高，两种作答反应时差异越大，固定参数迭代标准化残差法的优势越明显。使用该方法能够有效提高不努力作答识别的准确性。并且，基于该方法识别并替换为缺失的数据拟合模型，相比于基于原始数据拟合模型，能够有效减小参数估计结果的偏差。

第三，构建了评价固定参数迭代标准化残差法应用效果的判别指标，并探索了临界值，为实际中该方法的选择提供了建议。研究三基于迭代过程中标准化反应时残差的变化构建了判别指标。该指标能够预测使用固定参数迭代标准化残差法带来的被试参数估计误差减小的程度并指导方法选择。研究三通过模拟条件较细化的模拟研究，探索确定了判别指标的临界值。研究结果建立了判别指标使用流程，对应了四种处理方式：（1）不需要识别不努力作答，（2）使用非迭代方法识别不努力作答，（3）使用固定参数迭代标准化残差法识别不努力作答，（4）进一步检查测验或被试原因。判别指标及其临界值的使用为实际中固定参数迭代标准化残差法的高效应用提供了建议。

最后，研究四采用实证研究方式证明了本研究提出的固定参数迭代标准化残差法和其他方法在不努力作答识别结果，会聚效度，以及基于识别并替换为缺失后的数据拟合模型得到的参数估计结果等方面存在差异。并且，采用基于实际数据分布构造数据的方式，演示并说明了判别指标的使用。

综上，本研究提出了一种新的不努力作答识别方法——固定参数迭代标准化残差法及其判别指标。研究了该方法在不同模拟条件下的表现，结合该方法及其判别指标制定了数据处理时选择不努力作答识别方法的流程建议，具有一定的理论与实践意义。

In educational and psychological measurement, the primary goal is to obtain valid scores for students, which means that the information based on a test can purely reflect the latent traits of those students. Many traditional measurement models or methods (e.g., item response theory, IRT) are based on the assumption that examinees respond each item effortfully throughout the test. However, in practice, the prevalence of non-effortful responses from unmotivated participants has been repeatedly reported, which can be observed in either low-stakes or high-stakes testing situations (Bridgeman & Cline, 2004; Wise & Kong, 2005). There can be negative consequences if a standard IRT model is applied to data contaminated by non-effortful responses. Hence, it is vital to identify and delete non-effortful responses before data-analysis.

Recently, as computer-based testing becomes popular, it is easy to obtain information of many sources during testing (e.g., response time, denoted as RT). In order to ensure the accuracy of parameter estimation, a large number of methods to identify and reduce the effects of non-effort responses have been proposed based on RT. In the framework of traditional RT standard residual method, this study proposes a two-stage method, which is identifying effortful individuals to improve item parameter estimation, then fixing item parameters and using an iterative purification process. This method is denoted as CSRI (conditional estimate with fixed item parameters standard residual method using iterative purifying procedure) in this dissertation. The CSRI method, as well as the traditional method are compared under several simulation studies. Moreover, this study proposes a discriminative index to predict the improvement of person parameter estimation by using the CSRI method. Following the simulation studies in search of cutoff values for fit index in structural equation modeling, the optimal cutoff values of this discriminative index are found in this study as well. This study includes three simulation studies and one empirical application. The primary process and conclusions are summarized as follows.

First, in Study 1, a mixture model, using both response accuracy and RT information, to help differentiating non-effortful and effortful individuals and to improve item parameter estimation based on the effortful group is proposed and investigated. The mixture model applied in this study has several advantages, such as using all the information from response accuracy and RT, free of strong assumptions and so on. Two simulation studies are conducted to compare the proposed method against two existing methods (RT mixture model method and normative threshold 10 method with response behavior effort index < 0.8) in terms of classification of effortful/non-effortful individuals, and eventually the capability to improve item parameter estimation. In the first simulation scenario, the speed of all the examinees follows a normal distribution. In the second simulation scenario, the speed of all the examinees follows a mixed normal distribution. The manipulated factors are: sample size, test length, non-effort prevalence, non-effort severity, and the difference between RTs of non-effortful and effortful responses. The results show that the mixture model method can reduce the bias of item parameter estimates caused by non-effortful individuals, which exhibits more advantages when the non-effort severity is high or the RTs are not lognormally distributed.

Second, in Study 2, an iterative purification process based on a RT standard residual method to detect non-effortful responses is proposed and investigated. In this method, the item parameters are fixed to values estimated based on the effortful group identified by a mixture model method as in Study 1. In Study 2, only the first simulation scenario in Study 1 is considered. The manipulated factors are similar as in Study 1. The proposed method (CSRI) are compared with the noniterative method (conditional estimate with fixed item parameters standard residual method, CSR), as well as the traditional standard residual method in terms of classification accuracy and parameter recovery. The results show that CSRI leads to a much higher true positive rate with a small increase of false discovery rate when severity of non-effort is high. In addition, parameter estimation is significantly improved by the strategies of fixing item parameters and iteratively purifying.

Third, in Study 3, a discriminative index used to predict the relative reduction of RMSE (root mean square error) of person parameters and offer suggestions for method selection is proposed. The definition of this index is similar to SRMR (standardized root mean square residual) in SEM, which is computed based on the difference of the standard RT residuals in the first and second iteration. The cutoff values of this discriminative index are investigated under a large-scale simulation study by considering true positive rate and false negative rate simultaneously. Finally, the standard process of using this index is proposed, which includes four ways suggested to deal with the data: no need to identify non-effortful responses, using traditional noniterative methods to identify non-effortful responses, using CSRI to identify non-effortful responses, checking the testing process or examinees.

Last, an empirical study is also conducted to show the differences of these methods and illustrate how to use the proposed discriminative index.

In summary, this study proposes a new method to identify non-effortful responses (CSRI) and an index to offer suggestions to apply this method. The new method is investigated under several simulated conditions. Moreover, a standard process of cleaning data contaminated by non-effortful responses may be valuable for practitioners.