OLS regression requires minimal errors in the ordinate variable (x axis). Applying this in a situation where that condition is not met produces an erroneous estimation of the slope.

It is a commonly practiced error to use a scatter-plot to examine the relationship between two quantities. Performing an OLS regression in this context will be likely to under-estimate the result.

One way to check the magnitude of the problem is to repeat the OLS with the datasets reversed.

If the percentage error is comparable in both datasets, bisecting the angle gives a better estimation. However, detailed knowledge of the error, noise or other signal present may not be well known.

Advertisements