When and during what criteria do we use interaction variables in R for modelling? What if I run a lm() and it shows no variable as significant what do I do? If I have to use interaction variables how do I decide the interaction between the variables when I have around 30 columns. Can someone please suggest me a video/book to learn more about this topic, it will be really helpful.
 A: First of all, it is good practice to have prior hypotheses about the relationship between your predictors and your response variable before running your lm(). I don't know know if you did this before you ran the lm, but you might want to check if the response variable indeed has a linear relationship with most predictors. A lot of times linear relationship is assumed where might be log-linear or quadratic functions are better fits - but you are the best judge of your data.
Something else that you might want to check on is the correlation between your predictors. If there are large correlations between the 30 predictors, it can obscure the signal with the data.
If you do want to go down the interactions route, an interaction is basically saying that the effect of predictor A on the response changes with predictor B. These interactions are hard to conceptualise and harder to interpret, especially when there are a large number of predictors. The best way to do this is to approach this with a mechanistic understanding of your system. This will help to predict which interactions, if any might be important in your problem. Here is a similar question with some references in the answer.
