| 中文题名： | 基于 MCMC 算法的多维 IRT 模型在反应风格上的应用 |
| 姓名： | |
| 保密级别： | 公开 |
| 论文语种： | 中文 |
| 学科代码： | 04020005 |
| 学科专业： | |
| 学生类型： | 硕士 |
| 学位： | 教育学硕士 |
| 学位类型： | |
| 学位年度： | 2022 |
| 校区： | |
| 学院： | |
| 第一导师姓名： | |
| 第一导师单位： | |
| 提交日期： | 2022-06-19 |
| 答辩日期： | 2022-06-19 |
| 外文题名： | APPLICATION OF MULTIDIMENSIONAL IRT MODEL BASED ON MCMC ALGORITHM IN RESPONSE STYLE |
| 中文关键词： | |
| 外文关键词： | Midpoint response style ; Extreme response style ; MCMC algorithm ; PPMC ; MNRM ; IRTree |
| 中文摘要： |
李克特形式的自陈量表（self-report scale） 是一种用来测量个体的人格、 态度或价值观的工具， 在心理学、 组织行为学等诸多领域都得到了广泛地应用(Goldammer et al., 2020; Hong et al., 2020)。 然而， 受测者在作答这种形式的自陈量表时， 可能会倾向于选择或避免选择特定的选项。 受测者的这种作答倾向即为反应风格（ response styles）。 反应风格会使得测量结果出现系统性的偏差(McCrae, 2018; Wetzel & Carstensen, 2015a)。 目前， 研究者已开发了不同的前， 研究者已开发了不同的 IRT 统计模型来测量和控制反应风格。 然而， 如何进行模型选择以及这些模型在不同文化背景中的适用性的研究还很匮乏。 另外， 越来越多的研究者认为， 反应风格不仅仅可以被视为反应偏差， 还可以被视为稳定的特质， 其心理学意义有待进一步探索(Meiser et al., 2019a)。 为了解决这些问题， 本研究应用了原理不同的三种多维 IRT 模型对两种反应风格（中点反应风格、 极端反应风格） 进行了建模， 通过两个实证研究， 对不同模型的拟合结果进行了绝对比较和相对比较， 并在不同文化背景下探讨了受测者在作答量表时的认知过程， 在为 IRT 模型的选择和使用提供了可行性建议的同时， 也探讨了性别和年龄与反应风格的关系， 丰富了反应风格的心理学意义。研究一基于 MCMC 算法， 应用单维 IRT 模型 GRM、 多维模型 MNRM 和两个 IR 树模型（模型 APP 和模型 MPP）， 比较和探究了不同模型的拟合效果以及量表目标特质与反应风格的关系。 研究结果表明， 对于中国受测者来说， MNRM 的模型拟合结果是优于IR 树模型的， 并且绝对拟合可以作为相对拟合结果的有效补充。 从模型估计的目标特质来看， 三个多维模型估计的目标特质之间呈现了显著正相关。 从模型的估计的反应风格特质来看， 相较于 MNRM， IR 树的模型假设更符合受测者的真实作答过程， 并且可以识别出错误作答者。 研究二在研究一的基础上， 比较了不同文化背景的三个国家（中国、 俄罗斯和荷兰）的受测者在作答人格量表时认知过程的异同， 以及不同模型的适用范围。 另外， 研究二还从性别和年龄角度， 探讨了反应风格的心理意义。 结果发现， 不同国家的受测者在作答时的认知过程存在系统性的差异。 俄罗斯与荷兰受测者普遍更符合模型 APP； 中国的受测者则呈现了更为复杂的模式， 即在外向性、 宜人性、 尽责性三个维度上更符合模型MPP， 而在神经质和开放性这两个维度上更符合模型 APP。 另外， 研究还发现， 中国受测者具有更高水平的中点反应风格、 更低水平的极端反应风格。 从性别差异角度来看，不同性别的受测者之间反应风格差异很小。 从年龄角度来看， 极端反应风格和中点反应风格随着年龄的增长呈现出了非线性的变化趋势。
|
| 外文摘要： |
Likert scale is a tool used to measure an individual's personality, attitude or values, and has been widely used in psychology, organizational behavior and many other fields (Goldammer et al. , 2020; Hong et al., 2020). However, participants may tend to choose or avoid choosing certain options when responding to this form of self-report scale. This response tendencies are called response styles. Response styles can cause systematic bias in results (McCrae, 2018; Wetzel & Carstensen, 2015a). Researchers have developed different IRT models to measure and control for response styles. However, studies on how to choose the right model and the applicability of these models in different cultural contexts are scarce. In addition, more and more researchers believe that response styles can be regarded not only as systematic bias, but also as stable traits. Their psychological implications remain to be further explored (Meiser et al., 2019a).
In order to solve these problems, this study applied three multidimensional IRT models with different principles to construct two response styles (midpoint response style and extreme response style). Through two empirical studies, the fitting results of different models were compared. The absolute and relative comparisons were carried out, and the different cognitive processes of the participants when answering the scale were discussed in different cultural backgrounds. This study not only provides feasible suggestions for the selection and use of IRT models, but also explores the relationship between response style and demographic variables (gender and age), enriching the psychological significance of response style. Based on the MCMC algorithm, study 1 applied the one-dimensional IRT model GRM, the multidimensional model MNRM and two IRTree models（ APP and MPP） to compare and explore the fitting results of different models and the relationships among the scale target traits and response styles. The results showed that for Chinese participants, the model fitting result of MNRM was better than that of IRTree models, and the absolute fitting could be used as an effective supplement to the relative fitting results. There were significant positive correlations among the target traits estimated by the three multidimensional IRT models. Compared with MNRM, the model assumption of IRTree models was more in line with the real answering process of the participants, and could identify wrong answerers. On the basis of Study 1, Study 2 compared the cognitive processes of participants in three countries (China, Russia and Netherlands) with different cultural backgrounds when they responded to the personality scale. In addition, Study 2 also explored the psychological significance of response styles from the perspective of gender and age. The results found that Russian and Dutch participants were generally more in line with the model APP. The Chinese participants showed a more complex pattern, that is, they were more in line with the model MPP in the three dimensions of extroversion, agreeableness, and conscientiousness, while those in neuroticism and openness were more in line with the model APP. In addition, the study also found that Chinese subjects had a higher level of midpoint response style and a lower level of extreme response style. From the perspective of gender differences, there was little difference in response style between participants of different genders. From the perspective of age, extreme response styles and midpoint response styles showed a nonlinear change trend with age. |
| 参考文献总数： | 327 |
| 馆藏号： | 硕040200-05/22001 |
| 开放日期： | 2023-06-19 |
谷歌
