0

When I build a regression model, which is considered most important:

  1. removing insignificant variables,
  2. checking jfor multicollinearity and removing those variables that contribute to it,
  3. multiple R-Squared,
  4. or something else?

If I have to take care of all of the above, what should be my order of preference?

If I want to evaluate the model, is there is hard and fast rule that says residuals should be normally distributed?

Y.Surya Narayana
  • 31
  • What are your larger goals? What do you want from the model, once you've built it? – gung - Reinstate Monica May 16 '19 at 15:22
  • I am trying to predict a variable with my model. I did the back tracking and the results almost similar without one particular variable. By using min-max accuracy min_max_accuracy <- mean(apply(Backtrack, 1, min) / apply(Backtrack, 1, max)), I am getting values of 73.5 and if i take one extra variable, i am getting the value as 75. Which one should i consider ?. should i go with model with less variable ? – Y.Surya Narayana May 16 '19 at 15:25
  • OK. You can find a great deal of material about all this on our site. Here is a very focused search to consider. – whuber May 16 '19 at 15:29

0 Answers0