for an outcome (relapse after therapy, yes or no), when univariate logistic regression was applied; p value for each variable was significant, then if multivariate logistic regression was applied with enter method or forward selection, all the significant variables P values became high as 0.9 or even 1.0.
how can these results be solved and what is the explanation?, what step did we miss?
Check multicollinearity between the predictors .
Try to remove variables with weak relationship to the outcome.
If the sample size is small, consider reducing the number of predictors.
testing colinearity i think by VIP test in linear regression not binary logestic, despite this i tested all of the variables by VIP test and the value was less than 5
Testing collinearity can be applied in both linear and logistic regression and VIP for logistic regression (e.g. in SPSS) applied through collinearity in linear regression menues... while VIP value is less than 5, so multicollinearity is not an issue.. in this case sample size should be considered if small and other type of regression may be useful.. I heared about Lasso and Ridge to deal with small sample size.
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India