I fitted a linear regression with heteroscedescity using weighted least square method. How do I evaluate the quality of my fitting? I’m thinking by the p-value of slope/intercept. Or maybe because I minimized the cost function and that process already does the talking?

  • $\begingroup$ A p-value of a slope never measures the quality of a fit, at least not in any usual sense of "quality." Could you elaborate on what you mean by "quality"? $\endgroup$ – whuber Oct 29 '17 at 19:02
  • $\begingroup$ @whuber thats the question I would like to ask. How do I define the quality of the fit? $\endgroup$ – HanaKaze Oct 30 '17 at 21:11

A simple function to calculate R squared from a weighted LS fit would be

get_r2_ss <- function(y, y_pred, w) {
  # Calculate R2 using the Sum of Squares method
  # https://en.wikipedia.org/wiki/Coefficient_of_determination#Definitions
  ss_residual = sum(w * (y - y_pred) ^ 2)
  ss_total = sum(w * (y - weighted.mean(y, w)) ^ 2)
  return(1 - ss_residual / ss_total)

but if you use a weighted lm fit you will also simply get it using

summary(lm(y~x, weights=weights))$r.squared

If you want to look at the quality of the fit taking into account the nr of parameters that were fit then you can of course also use adjusted R squared

summary(lm(y~x, weights=weights))$adj.r.squared

or AIC or BIC, using

AIC(lm(y~x, weights=weights)) or BIC(lm(y~x, weights=weights))

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.