Skip to content

Models

2 messages · Val, Greg Snow

6 days later
#
If t1-t5 are all correlated with the outcome and with each other, than which are significant will depend on variations in the data (it is possible to have a set of values t1-t5 that predict the outcome well, but which all have nonsignificant p-values when taking the others into account).  Allowing gam to fit a non-linear function could easily lead to a different set being "significant".

My guess is that the question that you are interested in and the question that is answered by looking at the p-values are not as similar as you had hoped.  If you tell us what question you are trying to answer is, then we may be more helpful.