定序变量回归模型在心理数据分析中的应用

A Regression Analysis Model of Ordinal Variable to Psychological Data

定序变量在心理现象和心理数据中随处可见,采用综合的定序变量回归分析模型可以对"镜像模式"和"漏斗模型"的心理现象做出合理的解释和预测。首先通过非参数检验对影响因素进行初步降维,其次用Probit定序回归对降维后的影响因素贡献率进行判别,从而进一步筛选具有显著性判断水平的有效指标,最后用Logistic回归模型对某种特定的心理现象发生与否进行信息量足够大的解释和预测。大学毕业生工作生活质量满意度的预测对这种综合定序变量回归分析模型的实例拟合,证实了综合定序变量回归分析模型在心理现象和心理数据分析中的应用价值。

Ordinal variables are the common form of categorical variables in random phenomenon. Ordinal data which is formed from the level of ordinal variables by sequencing scale measurement has been widely used in psychological research. Psychological data is a kind of data from randomized hidden variable, which seems to be noticeable but could not be touched such as degree of satisfaction, preference degree, cognition degree, sentiment perceptibility, behavioral level and so on. The mental impression is hard to be calculated. To be exposed for calculated ordinal data is a kind of judgment standard or decision threshold criteria of an individual psychological activity in implicit psychological data. When a certain degree of psychological feeling happens to be just between two adjacent thresholds, the individual would be given a numerical value like a scale to project this ＂Mirror mode＂ of the psychological decision threshold criteria. Meanwhile, people are always concerned about what factors or conditions decide the high-low of threshold value of these ordinal variables based on cognitive instinct. This sort of ＂Hopper model＂ which is used to study the factors affecting to the psychological decision threshold criteria is a typical regression model. The paper proposes a multiple regression analysis model of ordinal variables based on the three problems involved in typical regression analysis ＂Hopper model＂ for ordinal variables. First, the ordinal variable indexes which can explain a psychological phenomenon are initially reduced in dimension by nor-parameter test method. Then, the effect indexes which have significant judgment standard are selected by using Probit ordinal regression. Finally, the probability of a psychological phenomenon happen is predicted and explained enormously by using Logistic regression model. Based on the data of quality of work life for the college graduates, the forecasting is suggested and the simulation is done for the ＂Mirror mode＂ and the ＂Hopper model＂ of ordinal variables regression model. It is hoped that ordinal variables regression analysis models and statistical analysis methods would have wider applied value in study of psychological phenomena included cognition, emotion, behavior, economy, social life etc.