Can you use logistic regression and linear regression methods interchangeably? If yes, in what scenarios?
Even if the target is binary, we can still use linear regression models if the importance of the model is just predicting the ranking estimates. Let us take an example where your manager wants you to find the top 20% or 30% of your customers based on who is likely to respond for an e-mail offer. You don’t need to predict who is going to respond and who is not going to respond. In this case, you need to rank order your customers with most probability of response to least probability of response and give your manager the top 20 or 30%. So in this case, y or the dependent variable is just the logit score or the ranking estimates. However, if the idea is to predict the decisions (i.e. classification of the predicted rankings into decisions) then we need the logistic regression model to calculate the prediction estimates based on the logistic function which uses the logit score. Then using an appropriate threshold value, these prediction estimates can be converted to decisions.