Objective The complexity of the concept of multivariate statistics can easily lead to difficulties in students'understanding.This article uses simulation experiments to examine the impact of various characteristics of data on the analysis re-sults,in order to strengthen students'understanding of the principles of multivariate statistics and facilitate their correct applica-tion of multivariate statistical methods.Methods Six simulation experiments were designed,and R 4.2.2 was used to generate simulation data and conduct statistical analysis to explore the impact of data characteristics on the 6 statistical analysis methods.Results According to the statistical analysis results of simulation experiments,the multicollinearity among independent variables in multilinear regression will make the regression coefficient estimation of the regression equation inaccurate.The correlation structure between variables in principal component and factor analysis affects the dimensionality reduction effect.In logistic re-gression,the nonlinear influence of independent variables on dependent variables should be considered,otherwise the results will affect the accuracy of the estimation results.When the proportion of deletion in Cox regression is too large,especially after reac-hing 60％,the results of the Cox regression model have a significant deviation.The correlation structure between variables affects the results of canonical correlation analysis.When there is a large difference in the proportion of categories,Bayesian discrimina-tion and logistic regression are more appropriate for discrimination,because Bayesian discrimination with prior probability is more flexible.Conclusion The application of multivariate statistical methods needs to fully consider applicable conditions such as data characteristics.Through simulation experiments,students are encouraged to independently run simulation experiments,intuitively understand abstract multivariate concepts,and facilitate the understanding and application of multivariate statistical methods.
维普
万方数据