基于人工神经网络的一种效度凭证求取方法

针对基于统计学的效度凭证求取方法所存在的问题，提出了基于人工神经网络的效度凭证求取方法。蒙特卡罗模拟实验和对实际数据的分析表明，当测验分数和效标分数为单变量且非线性关系时，或者测验分数和效标分数为多变量时，运用神经网络方法可以比统计学方法更好地求取心理测验效度的基于和其他变量关系的凭证。

The Study on Artificial Neural Networks Based Validity Argument Obtain Method

Because there were some shortcomings that validity argument was obtained with statistics. The artificial neural networks based validity argument obtain method was introduced. 21 Monte Carlo simulation experiments were conducted, including the relationship between single independent variable and dependent variable were linear or nonlinear, the relationship between multi-independent variable and dependent variable were linear or nonlinear, the relationship between independent and dependent variable were random, the relationship between variables were affected by random variable with different degrees. Both standard normal distribution and uniform distribution for the independent variables were in each experiment. The actual psychological measurement data were analyzed with statistics and artificial neural networks. The resuits showed that the neural networks method was better than the statistical method for obtain the validity argument, when the relationship between the test scores and criterion scores was nonlinear or the test scores and criterion scores were multivariables.