心理科学 ›› 2011, Vol. 34 ›› Issue (6): 1482-1487.

• 统计与测量 • 上一篇    下一篇

等级数据的测量不变性检验及影响因素模拟研究

李冲,刘红云   

  1. 北京师范大学
  • 收稿日期:2010-09-13 修回日期:2011-04-27 出版日期:2011-11-20 发布日期:2011-11-20
  • 通讯作者: 刘红云
  • 基金资助:

    国家自然科学基金青年基金项目

Simulation study of the detection of measurement invariance and its influential factors for ordinal categorical data

Chong LI,   

  • Received:2010-09-13 Revised:2011-04-27 Online:2011-11-20 Published:2011-11-20

摘要:

研究介绍了针对等级数据的模型建构(LRV,潜在反应变量模型)和参数估计(WLSMV)方法,以及在此基础上的测量不变性检验(DIFFTEST)方法,同时采用蒙特卡洛模拟研究方法,考察样本总量大小、组间样本量对比情况、阈值差异程度、量表长度等因素,对DIFTEST进行针对等级数据的测量不变性检验效果的影响情况,以及WLSMV估计方法下的参数复原情况。研究结果发现WLSMV估计方法参数的复原效果很好;DIFFTEST的一类错误概率达到可接受水平,在大样本情况下、组间样本量基本相等、阈值差异程度较大时,DIFFTEST检测力较好。在控制测量不变性遭受破坏的项目个数情况下,随着测验长度的增加,DIFFTEST的检测力下降。

关键词: 等级数据, DIFFTEST, 测量不变性, 模拟研究

Abstract:

The study introduces a new method of model construction and parameters estimation for ordinal categorical data,with the method of exploring measurement invariance based on this method. simulation study explores the effect of parameters recovery, difference between estimated value and known population parameters value ,empirical power under various total sample size, the ratio of reference group to focal group size, the difference of parameters between two groups, test length. Results indicate that the parameters recovery well when robust weighted least squares estimator using a diagonal weight matrix(WLSMV) are used. The type Ⅰerror of DIFFTEST consistently adhered closely to the nominal alpha level of 0.05.Under the condition of large sample size, reference and focal group size are equal, the difference of parameters is bigger, the power of DIFFTEST show better. Control the number of items whose measurement invariance are violated, the power of DIFFTEST become lower when the test length become longer.

Key words: Ordinal Data, DIFFTEST, Measurement Invariance, Simulation Study