Abstract: Psychological data is often obtained from subjective assessment，it has larger error compared with physical data.Creativity measurement data is applied in this study.Generalizability theory is used to describe the quality of psychological data.The model of psychological data is set up by BP-Adaboost RT method，and compared with the multi-variable linear regression model and single BP network model.The results show that generalizability theory can be used to analyze psychological data，and the data quality will affect the predictive accuracy of the model.

Key words: data quality, generalizability theory, Adaboost, BP neural networks, creativity measurement

摘要： 心理学的数据往往是通过主观评判得到的，和物理学的数据相比有较大的误差。以创造力测量为例，运用概化理论对心理测量数据的质量进行了描述。采用BP-Adaboost RT方法对心理测量数据进行建模，并且和多元线性回归、单个BP网络的建模进行了比较，结果表明，概化理论可以用来分析心理测量数据，数据的质量会影响到模型的预测精度。

关键词: 数据质量, 概化理论, Adaboost, BP神经网络, 创造力测量