Basic question about stepwise regresson

#1
Hello everyone, total stats beginner here so I apologize in advance if my question is silly.

I did a stats assignment recently where we had to build regression models. I started by building a model where I added all of the explanatory variables to the model, and then I used a backward/forward stepwise regression with BIC criterion to help me choose the best model. One thing that happened that I couldn't understand was a few explanatory variables that were statistically significant in the model with all of the explanatory variables in it were left out of the model with the lowest BIC. In addition, a few variables that were not significant in the initial model became significant in the model with the lowest BIC.

My tutor said that "variables that are not significant when other variables are controlled for may become significant when you stop controlling for them."

Can anyone elaborate on what he meant by that?

Thanks for any insight, it's greatly appreciated.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
In actual practice stepwise regressions are considered garbage/joke. They do not take into account context knowledge. If you had confounders, mediators, common effects of exposures and outcomes, or your IV and DV flipped - stepwise doesn't know this - since it is just an automated process. Ideally you map out ahead of time how all of the variables are related prior to selecting terms to be included in a model. Any of the underlined issues above could be a culprit.
 
#3
In actual practice stepwise regressions are considered garbage/joke. They do not take into account context knowledge. If you had confounders, mediators, common effects of exposures and outcomes, or your IV and DV flipped - stepwise doesn't know this - since it is just an automated process. Ideally you map out ahead of time how all of the variables are related prior to selecting terms to be included in a model. Any of the underlined issues above could be a culprit.
Thank you very much hlsmith. Really appreciate that insight.