PSYCH OpenIR  > 健康与遗传心理学研究室
人格测验中自比数据对潜结构探测的影响
其他题名How the Ipsative Data Impact the Latent Structure Detection in Personality Background?
任芬
学位类型博士
导师张建新
2015-05
学位授予单位中国科学院研究生院
学位授予地点北京
学位专业心理学
关键词单一模式 迫选模式 自比数据 分类测量学 多维 IRT
摘要单一刺激模式的测验在整个测验发展史上统治了很长一段时间,这种测验的计分方式导致了很多测验偏差的发生(如,社会赞许性等)。为了避免这些反应偏差的出现,迫选模式的测验出现了,它可以在节省测验时间的基础上有效地降低单一刺激模式下的反应偏差。但是,由于传统的计分方式,迫选测验的计分会导致出现自比数据(Ipsative data)。
本文的研究目的是在模拟和实证数据的基础上, 使用分类测量学(Taxometric)的三种常用程序考察自比数据的传统和 IRT 处理方式对潜变量属性探测的影响。本文一共包括5个研究。
研究一采用 Ruscio  和  Kaczetow 基于 R 语言的迭代技术进行所有的数据模拟,分别模拟了单维和多维的数据,模拟的样本量为 500,变量为12个,重复次数为100 次。此外,使用同样的技术结合变量分割方法模拟了类别数据,类别数据的模拟考虑了类别之间的距离和基础类别比率。结果显示, 对于单维数据来说,自比数据的性质和区组大小不会影响潜在变量的探测; 多维数据的情况相对比较复杂,甚至有被误判为类别的情况出现,维度的增加和区组大小对 MAMBAC 程序的影响大于对 MAXEIG 和 LMode 程序的影响。MAXEIG 程序需要合适的区组大小才能保证不受影响。潜结构是区分较大的类别时,MAMBAC 程序的表现最为稳健,不受基准概率和区组大小的影响;MAXEIG与 LMode程序受到基准概率的影响但不受区组大小的影响。潜结构是区分较小类别时,虽然各指标有变化,但是偏差均在可接受范围之内,自比数据不会影响潜在结构的探测。
研究二主要以实证数据考察人格的可能模型对潜在变量探测的影响。样本容量为520 名在校大学生,来自贵州省和北京市的两所高校。在拟合模型之前,为了确保现有的样本量对于可能的模型是足够的,首先进行了基于人格大五模型的样本量合适性的模拟,结果表明现有样本量 520 人对于大五人格模型是足够的。所有可能的5个验证性模型的比较发现 MIRT 模型最拟合现在的数据,使用各个 模型得到的潜变量得分作为指标进行 Taxometric 分析,结果显示 MAMBAC 的结果是稳定的,其余的 2个程序表现出现了偏差。同样对于迫选模式的原始数据拟合可能的 5个模型,发现迫选模式的验证性因素分析全部都没有收敛。结果表明,自比数据的原始反应和传统 01 计分方法对 Taxometric 分析的影响较大,数据的潜在结构甚至被认为是类别的。
研究三采用 Thurstone  IRT 模型的计分方式对自比数据进行处理,以潜在因子分作为分析指标,结果明确提示潜在变量的维度属性。潜在变量之间的相关系数从严重扭曲恢复到正常状态。
研究四引入个人拟合指标和题目拟合指标 zh 对各个模型的不拟合的被试和不拟合的题目进行探测,结果显示各个模型使用 zh指标均得不到理想的结果。
研究五对自比数据的 IRT处理方法下,进行问卷信度和效度的估计,结果显示传统的 IRT 处理方法和 Thurstone  IRT 处理方法信效度系数均达到了良好的水平。
其他摘要As an usually used item presenting format, single item format has been used for a long time in the history of measurement. But this kind of format has many pitfalls in practice. In order to solve those problems(such as, social desirability), forced choiced format showed up and can reduce the impact of kinds of response biases. However, scored with traditional method, this kind of present format can produce ipsative data, which has psychometric problems, even can torture the instinct of the latent structure.  
Statistical analyses investigating latent structure can be divided into those  that estimate structural model parameters and those that detect the structural model type. The most basic distinction among structure types is between categorical (discrete) and dimensional (continuous) models. The taxometric method was developed specifically to distinguish between dimensional and 2-class models. Using the comparison curve fit index across 3  taxometric procedures (Mean Above Minus Below A Cut, Maximum Eigenvalue, and Latent Mode Factor Analysis) as the criterion for latent structure.
There are five studies  in this dissertation.  In study 1, with simulation data sets, under different conditions, we tested how the number of dimensions, the block size, the distance of taxon, and the base rate can impact the detection results and CCFIs.  It is implied that when the latent structure is dimensional, with recommended block size and number of dimensions,  the forced choice format would not  influence the  taxometric results. When the latent structure is categorical, the validity of indicators and base rate cannot impact the taxometric procedures.
In study 2, we took impirical data to investigate different classical scoring methods in five possible personality models. Five hundred and twenty students from Guizhou province and  Beijing city completed the chinese version of the big five personlity questionnaire in paper and pencil formats in return for a feedback report. Out of 520 participants, 203 were male and 296 were female,the others are missing values. Age ranged from 19 to 24 years, with a mean of 21.14 and a standard deviation of 1.39 years. The single-stimulus format was followed by the forced choice counterpart,where the participants were required to make choices in the most-least like format out of the same block of 4 items. The items in the same block are from different dimensions, measured different traits. Two formats were used for compare consideration. In this paper, we also used simulation method to check the current sample size is enough for the confirmative models.  Among all the five possible models for single-stimulus personality questionnaire, multidimentional item response theory(MIRT) fitted better than others. Computing the correspond latent factor scores as indicators in taxonmetrica analysis. But all models for the forced choice format raw data cannot  convergent. Then, we recoded the raw forced choice data and analysised in MIRT framework. The raw and classical scoring methods for ipsative data had different effects  on  taxonmetrica procedures.  
As for study 3, Thurstone  IRT model was used as a new scoring method to fit ipsative data. The new method corrected the correlation matrix and clearly indicated the latent structure is dimensional.
In study 4 of  this research, we also  took  the same empirical big five personality data  as in study 2  to test  if  the MIRT model can  detect the bad items and  aberrant responses.  The index used in the comparation  are  item fit and person fit  zh. Unfortunately, zh index performed badly in ipsative data even scoring in Thurstone IRT model.  
The last study, the reliability and standard error of measuement were caculated. The IRT empirical reliabilities are higher than alphas for the classical ipsative scores. Both the conclusion and limitations of these results are discussed.
学科领域医学心理学
语种中文
文献类型学位论文
条目标识符http://ir.psych.ac.cn/handle/311026/19652
专题健康与遗传心理学研究室
作者单位中国科学院心理研究所
推荐引用方式
GB/T 7714
任芬. 人格测验中自比数据对潜结构探测的影响[D]. 北京. 中国科学院研究生院,2015.
条目包含的文件
文件名称/大小 文献类型 版本类型 开放类型 使用许可
任芬-博士学位论文.pdf(3068KB)学位论文 限制开放CC BY-NC-SA浏览 请求全文
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[任芬]的文章
百度学术
百度学术中相似的文章
[任芬]的文章
必应学术
必应学术中相似的文章
[任芬]的文章
相关权益政策
暂无数据
收藏/分享
文件名: 任芬-博士学位论文.pdf
格式: Adobe PDF
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。