6
$\begingroup$

I have a question as a newbie. I'm working on a tool using regression analysis( linear, multiple, multivariate) to derive a regression model. To verify the correctness of the tool, I'm trying to find a benchmark, or a benchmark dataset to test the tool. Any idea or comments are all welcomed. Thanks in advance.

$\endgroup$
1
5
$\begingroup$

See NIST's Statistical Reference Datasets. These include data chosen or designed to present numerical challenges to regression algorithms. The Longley data, with highly collinear predictors, is perhaps the most famous example.

$\endgroup$
1
  • $\begingroup$ That's a good start thanks! Any ideas where to get a dataset with more observations? $\endgroup$
    – Verena
    Sep 2 '16 at 7:32
4
$\begingroup$

Generate some random data yourself in whatever language you're using, that follow the assumptions of your model, e.g., for linear regression generate $X$, then $\beta$, then do $y = X\beta + \epsilon$ where $\epsilon$ is normally distributed with mean zero and sd of say 1. See if you can recover the correct $\beta$ when varying the error stdev. Compare with the multitude of established tools for doing regression.

$\endgroup$
2
  • 1
    $\begingroup$ (+1) Certainly useful first steps, but you'd want some hard test cases if possible. $\endgroup$ Jun 10 '15 at 18:37
  • $\begingroup$ Simulation is always good but if you want to publish an algorithm you want to have comparable results to others. $\endgroup$
    – Verena
    Sep 2 '16 at 7:30

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.