Statistic test when bias is not random

Question

Is there a statistical test that I can use to say that there is bias in the result of my analysis but the bias only occurs at certain region. The scatter plot below probably will make more sense:

Basically this are the residuals vs observed data. I hope you can see that as the observed data increases, the residual increases. The overall result was unbiased calculated using mean error, but clearly something is wrong at the higher end. Hope to get some ideas. Thanks

Answer 1

Testing residuals is not common practice (see for example) R: test normality of residuals of linear model - which residuals to use) In this case your plot could be an indication of non-linearity, did you try to add an quadratic component in your model (assuming it is a regression model)? Basically you square the data and add it to the model.