不同定义平行测验等值的群体不变性

群体不变性是等值的一个重要假设,即对不同的考生子群体等值函数一致。本研究对不同平行测验定义下线性等值的群体不变性进行了理论分析和模拟研究,模拟研究REMSD指标通过六种不同加权方式计算。结果显示,严格平行测验在信度较低时REMSD指标更大;子群体均值差异和信度差异对REMSD的影响存在明显的交互作用;REMSD指标在期望权重等权下的最大,在分数权重采用子群体比例加权最小。最后对结果进行了讨论,对REMSD权重使用及进一步研究给出了建议。

Population Invariance of Linear Equating under Different Definition of Parallel Test

When constructing the parallel tests to measure to the same construct,it must ensure the comparability of test scores across different test forms by equating.Population Invariance is one of the important assumptions in equating that the different equating functions based on subgroups of candidates should be consistent with each other.In the present paper,population invariance of linear equating is investigated under different definitions of parallel tests by theoretical analysis and simulated data and REMSD is calculated by six different ways of weighting.The results showed that,inconsistent with previous results,REMSD is high when reliability of strictly parallel test is low;differences in subgroup means and reliability shows interaction;REMSD is the largest under weighting of expectation equally,and smallest under weighting of score by subgroup proportion.Discussion of results is followed and suggestions for weight in REMSD and further research on population invariance are also given.