I have a dataset with known errors in both the X and Y and want to perform a simple linear regression. From reading other posts, it seems I want to TLS over OLS due to the presence of error in both variables. However, the error is no constant, so I also want to weight measurements.

It seems that weighting with the inverse of the variance is common in weighted OLS. Could I weight with the inverse of the sum of X and Y variance for my TLS regression?

  • 1
    $\begingroup$ So the standard weighting procedure when you have errors in both uses a weighted sum of the inverses: en.wikipedia.org/wiki/Total_least_squares $\endgroup$
    – VCG
    Aug 19, 2016 at 21:09
  • $\begingroup$ Great thanks! Should have checked Wikipedia first I guess $\endgroup$
    – rconway91
    Aug 19, 2016 at 23:46


Your Answer

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