Lord长空一剑 发表于 2010-3-7 07:41
you may manipulate the variables if it's insignificant.
you've mentioned that e variables are significant individually, but only one maintains its significance when all the variabled combined.
hence, it is true that only one variable should be taken when you have concerned all the factors listed together in this model.
however, you 'll have to argue why .
you have to because the two wre significant, but when they act with other factors , they are difersified away. so there must some interlinks between these variables.
you may first reason it using your economic sence, because econometrics is economics on the earth, statistical model is only a tool. you have to interpret the economic meaning of it.
then , mathemetically and statistically , you can have a look at the correlation between each variables. you will find the reason from it. if the correlations are negtive between some variables , then it is easy to explain why these two variables, which were initially significant in each individual models , are driven away from the whole model.
When i analyse the control variables individually, there are three significant ones.....DR ER FS. and then when i put them together, there is only one left. that is FS. and then i found DR and ER have collinearity....i used correlation matrix table to analyse collinearity.
I was wonderring, when variables have collinearity, i do not need to delete them..right? the book said sometimes after deleting, maybe new problems come up...so should i understand it as i can keep them and do nothing...or i should do something to deal with collinearity?
i have a question:
can i keep DR and ER? And for other control variables that are always insignificant, should i delete them or else?