liuzhen529
commented
6 years ago Hi Kelly,
Thanks for your effort for our project. I am not sure which criteria you used and how you write your code. For my understanding, for example, when using forward-stepwise AIC, like what I wrote in our project:
"step(null, scope = list(lower=null,upper=full), direction="forward",criterion='AIC')" means adding the variable one by one from the starting model. 'Scope' restricts the range and the starting model is 'null'(object).
I guess that if you just changed direction to "backward", which means decreasing the variable one by one from the starting model, the result must be intercept since the starting model is still null and nothing can be decreased. If you want to try "backward", the object and scope probably need to be changed.
Hope my words would help you :)
Sincerely, Zhen
KellyHu
Hi everyone!
I wonder how to explain which direction we pick for our project? I tried direction="both", "forward" and "backward". It turned out that the results were the same for "both" and "forward", but only the intercept was chosen when it was "backward". I think we might go with "both" or "forward". However, how to explain we prefer "both" to "forward" (or "forward" to "both")?
Besides, BIC chose only 2 explanatory variables, which was far less than the number of variables chosen from AIC. Shall we still keep our results for BIC? Then how should we conclude in the end?
Our group members deal with multiple variables and try different methods for our project. Hence, we already got hundreds of lines of code. Do you have any advice on what to put in our report to make it more concise? Is it ok to pick just one method and go deeper? Otherwise, the report will be too long...
Thanks!