果果250 发表于 2010-4-24 15:14 
偶目前在写一篇文章，其中有一部分内容是要从十几个自变量中找出对因变量有显著影响的变量。老师让我根据前人的研究成果或是自己的思考，在十几个变量中挑选几个变量成一组合做回归分析，然后再挑选几个变量成一组合做回归分析，这样子挑选出十几个组合各自进行回归分析后，然后比较这些回归的拟合程度，拟合程度好的组合里面的变量就认为是有显著影响的变量。我怎么感觉这样的分析思路我以前从来没有听说过，不知各位计量达人听说过没有，认为这样的思路科学严谨吗？
Usually the econometric model is based on economic theory--which is different from statistic model. But some time it is very difficulty to differencial the measurements in reality. For example, the theroy says INCOME is one of the important facters. The question in reality is which sort of income should be used personal income, disposable income, etc.
Also it is important to understand the difference between forecast model and policy evaluation model. We would like a model is good for both. But often in reality, a model is good in one aspect. In forecast models, the forecast accuracy is important while the unbiasedness is important in policy evaluation models. To select model based on "拟合程度" only is not a good idea.
Based on what you said, your teacher 思路 BO 严谨.
扫码加好友，拉您进群
全部版块
我的主页
收藏
