1
$\begingroup$

I was reading through my stat book and it was written that bootstrapping can relax the distribution assumptions for linear regression generalizability. I do not quite understand what assumptions we might have for linear regression distribution (maybe have something about residuals).

I am attaching the screenshot(book textparagraph about bootstrapping for robust linear regression), it would be great if anyone can shed some light on this topic.

Improve this question
$\endgroup$
1
  • $\begingroup$ See sec 7.7.2.1, of the same book, in particular, p272. The most commonly used tests and confidence intervals and prediction intervals for regression all assume normality of errors. $\endgroup$
    – Glen_b
    Commented Apr 4, 2020 at 5:32

1 Answer 1

Reset to default
2
$\begingroup$

One general use of bootstrapping (not the only one) is to estimate parameters and their p values, confidence intervals and so on when there is no analytic solution.

The usual methods of getting p values and CIs in OLS regression rely on analytic methods. These analytic methods assume that the errors are normally distributed. But bootstrapping does not assume this.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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