1. Before fitting a regression model, please explore the data, to look for trends and outliers.
2. Test for multicollinearity in the variables.
3. Test for interaction terms.
4. Perhaps in nonlinear function, polynomial funcions or discrete version some variables migh become significant, so you might use smoothing techniques to look for it.
5. Removing outliers and refit the regression model.
6. Exam the model assumptions.

首先检查模型设定是否有问题；其次按显著性由低到高逐个删除，直到达到研究要求。。。

3# zhangchx

Which variables will be kept in the model and which variables will be removed from the model should depend on the research questions.

sometimes, even variables with very large p-value must be kept in the model if it is of interest according to the experimental design.

If there no specific restriction, most of the time, we do backward selection, forward selection or hybird selection.

做一下回归诊断看看?比如,共线性诊断.

逐步回归