2

I am struggling with this problem, can anyone tell me how can I choose the most important variables among several independent variables before fitting any models in GLMs? I really appreciate it. I have already fitted some models but it doesn't make sense for me. Thanks in advance

kjetil b halvorsen
  • 77,844
Alan
  • 21
  • 3
    Welcome to the site, @Alan. Your question is a little sparse to be answerable. Would you mind telling us more about your situation, your data, & your goals? Since you're new here, it would help you to read our about page & our FAQ, which contain info about the site & how to ask questions. You might also find this blog post helpful in formulating your question: how to ask a statistics question. – gung - Reinstate Monica Feb 20 '13 at 00:37
  • 2
    In general, choosing important variables with which to fit a model is done by thinking about the nature of the situation & what prior theory dictates is relevant. – gung - Reinstate Monica Feb 20 '13 at 00:39
  • 1
    Well, first thanks for replying, I have more than 20 independent variables, I need to find the model which fits the data well. So, in this case, I want to find a way which variable has the most effect on the response variable in order to put into the model. Is it helpful?? – Alan Feb 20 '13 at 00:55
  • For the most part, that is an extremely difficult problem. You may want to read my answer here: algorithms-for-automatic-model-selection. – gung - Reinstate Monica Feb 20 '13 at 01:24
  • 2
    Hi Alan. As @gung said, this is a hard problem. In your reply to him, you say you "need a model which fits the data well" that is not the same as finding the most important variables.
    In many fields of research, many models will fit the data well and it can be very hard (if possible) to say which fits it **best**. Getting a better answer than that would depend on your needs; see gung 's first comment and tell us a lot more and then we can help more.     – Peter Flom Feb 20 '13 at 12:21
  • I understand what both of you commented, but the point is I have to put one of them into the model right? For example, for a particular disease if my independent variables are (Age,Gender,Smoke, Exercise and so on. I was asked to fit the models like this(model1<- glm(diseas ~1, family=poisson)) and the second model is( model2<- glm(diseas ~ Age,family=poisson)) as you we need to compare these two and decide whether Age is significant or not. My problem is when I accept the second model and after taking forward selection, I may end up with the best one, but if I put Gender first time, – Alan Feb 20 '13 at 21:00
  • The model would be another one. The whole thing is how do I know which one must be put into the model as a first variable?? I hope you can understand me. – Alan Feb 20 '13 at 21:01

0 Answers0